Zelinskywibbelt C Ed The Semantics of Prepositions From Ment

The Semantics of Prepositions
Natural Language Processing

3
Editorial Board
Hans-Jrgen Eikmeyer
Maurice Gross
Walther von Hahn
James Kilbury
Bente Maegaard
Dieter Metzing
Makoto Nagao
Helmut Schnelle
Petr Sgall
Harold Somers
Hans Uszkoreit
Antonio Zampolli
Managing Editor
Annely Rothkegel
Mouton de Gruyter
Berlin New York
The Semantics of Prepositions

From Mental Processing to
Natural Language Processing
Edited by
Cornelia Zelinsky-Wibbelt
Mouton de Gruyter
Berlin New York 1993
Mouton de Gruyter (formerly Mouton, The Hague)

is a Division of Walter de Gruyter & Co., Berlin.
Printed on acid-free paper which falls within the guidelines

of the ANSI to ensure permanence and durability.
Library of Congress Cataloging-in-Publication
Data
The Semantics of prepositions : from mental processing to natural

language processing / edited by Cornelia Zelinsky-Wibbelt,
p.
cm. (Natural language processing ; 3)
Most papers were presented at a workshop held Feb. 1990 at the
Institut fr Angewandte Informationsforschung.
Includes bibliographical references and index.
ISBN 3-11-013634-1
1. Grammar, Comparative and general Prepositions Congresses.
2. Semantics Congresses.
3. Cognitive grammar
Congresses.
4. Computational linguistics Congresses.
I. Zelinsky-Wibbelt, Cornelia. 1 9 5 2 II. Series.
P285.S46 1994
415 dc20
93-30213
Die Deutsche Bibliothek Cataloging-in-Publication
Data
The semantics of prepositions : from mental processing to

natural language processing / ed. by Cornelia Zelinsky-Wibbelt.
Berlin ; New York : Mouton de Gruyter, 1993
(Natural language processing ; 3)
ISBN 3-11-013634-1
NE: Zelinsky-Wibbelt, Cornelia [Hrsg.]; GT
Copyright 1993 by Walter de Gruyter & Co., D-10785 Berlin.

All rights reserved, including those of translation into foreign languages. No part of this
book may be reproduced or transmitted in any form or by any means, electronic or
mechanical, including photocopy, recording, or any information storage and retrieval
system, without permission in writing from the publisher.
Typesetting: Arthur Collignon GmbH, Berlin. Printing: Gerike GmbH, Berlin.
Binding: Lderitz & Bauer GmbH, Berlin.
Printed in Germany
Preface
I wish to express my thanks to all authors who have contributed to this

book. The majority of the papers in this volume were presented at a
workshop on the semantics of prepositions held at the Institut fr Angewandte Informationsforschung in February 1990. The range of the book,
however, has been enlarged by the contributions of several colleagues
who could not participate in the original workshop, but who kindly
complied with the editor's request for further submissions.
On behalf of all authors, I should like to express my gratitude to the
following institutions and individuals for their role in the production of
the present volume. The financial prerequisites for the holding of the
workshop were provided by funding from the BMFT (Bundesministerium
fr Forschung und Technologie, Frderkennzeichen 08B3213 8). The
encouragement of the project manager Johann Haller was an essential
factor in getting the workshop off the ground and in ensuring that the
papers were able to be published in a common volume. Finally, I should
like to acknowledge the assistance of my husband, who, in addition to
his own busy schedule, helped with the compilation of the index.
The broader topic of this book is the conceptualization of space. It
was the editor's intention to shed some light on this topic by bringing
together different approaches to the mental process of interpreting prepositions. In particular, the editor wanted to open windows for the computational processing of prepositions as predications in different cognitive
domains. While it was a complex matter to get all these different perspectives under one heading, the editor is aware that it is by far not the
whole area of current research on prepositions that has been covered.
No presentation of a subject, however, can ever claim to be exhaustive,
and there will always be room left for other perspectives.
In this vein we hope that this book, with all its valuable insights as
well as its inevitable flaws and gaps, will inspire future discussions on a
major topic within Cognitive Linguistics.
Saarbrcken, July 1993
Contents
Introduction
1. Lexicalization patterns of prepositions

The Dutch spatial preposition "in": A cognitive-semantic analysis
Hubert Cuyckens
27
Dividing up physical and mental space into conceptual categories

by means of English prepositions
Rene Dirven
73
On the grammar of lexical and non-lexical prepositions in English

Gisa Rauh
99
Prepositions:
patterns of polysemization and strategies of disambiguation
John R. Taylor
151
2. Semantic categorization of prepositions and context

Prepositions and object concepts:
A contribution to cognitive semantics
Priska-Monika Hottenroth
179
Semantic and conceptual aspects of the preposition durch

Ingrid Kaufmann
221
The meaning of German projective prepositions:

A two-level approach
Ewald Lang
249
3. Contrastive implications of prepositions

German temporal prepositions from an English perspective
Martin Durrell David Bree
295
On universality and variability in the semantics of spatial adpositions

Bruce W. Hawkins
327
viii
Contents
Interpreting and translating prepositions:

A cognitively based formalization
351
4. Image understanding and prepositions

A three-level approach to the semantics of space
Michel Aurnague Laure Vieu
393
Computing the meaning of localization expressions involving

prepositions: The role of concepts and spatial context
Simone Pribbenow
441
A contribution to reference semantics of spatial prepositions:

The visualization problem and its solution in VITRA
Jrg R. J. Schirra
471
Index
517
Introduction 1
Cornelia Zelinsky-
Wibbelt
1. The linguistic relevance of prepositions

What makes up the meaning of words has long concerned linguists both
from a general and from a typological point of view. In order to make
the necessary semantic distinctions between different meanings of a word,
an adequate typing in terms of semantic features is indispensable. Without
being able to make these distinctions, natural language analysis, synthesis
and generation as well as machine translation will be futile. One aim of
this book is to show which types of information are necessary in order
to arrive at these distinctions. While all authors achieve this aim by
proceeding from the idea of the human mind as an information processing
system, they differ with respect to their assumptions of necessary types
of information and possible uses of a word as well as their computation.
This makes it clear that the achievement of this aim is far from being
trivial.
This volume deals with the semantics of a lexical category which had
long been neglected in linguistic inquiry. Recently, however, interest in
prepositions has grown steadily, such that now we are witnessing a
veritable plethora of investigations into the semantics of prepositions. So
now seems a good time to reconsider this area of investigation. It is the
intention of this volume to approach this issue firstly from a more general
perspective, namely: how far do insights into the meanings of prepositions
give clues to the semantic structure of lexical units and its processing in
general? We also feel, however, that the time has come to see how natural
language processing (NLP) can benefit from the insights of theoretical
linguistics, especially with respect to machine translation (MT) and image
understanding. We think that it is particularly important to deal with the
meanings of prepositions from precisely these two perspectives: M T
requires a model of translation which takes into account the relation
between potentially different encodings of one and the same state of
affairs. This phenomenon is especially characteristic of prepositions, as
we will see below. Many of the articles in this book offer a model for
interpreting the typological differences in the distribution of information
in general. Both MT and the interpretation and linguistic encoding of

visual information require a comprehensive model of what knowledge
resources lie behind the use of linguistic expressions and which processes
operate on the knowledge resources in order to instantiate a specific use
of linguistic expressions. Certain techniques for interpreting the real world
can be explored by investigating the kind of information that prepositions
encode. Only a model of natural language processing can give valid
insights into the information encoded by lexical units. Different linguistic
theories arrive at different models by assuming different techniques for
understanding the real world. Among the most important of these is the
technique of abstracting from the vast amount of highly complex and
variegated information, which would exceed the processing and storage
capacities of the human mind. Thus the human mind, as an information
processing system, reduces this abundant information by focussing on
those aspects which, according to conventional experience, figure as
salient, essential or relevant in a particular type of situation, as well as
from the speaker's perspective. This focussing implies that the remaining
information is neglected as being non-salient, non-essential or non-relevant in the respective type of situation. With prepositions this abstraction
of information can be studied in a fairly transparent way: in their
prototypical meaning most prepositions are spatial predicates and as such
they can focus on different aspects of our physical environment. In fact,
the selection from among different prepositions or from among different
senses of a preposition depends on the possibility of focussing on a state
of affairs in different ways.
2. New ways of analyzing prepositions

The recent interest in prepositions is connected with the emergence of a
new philosophical climate in linguistics. The radically positivist and realist
climate which still dominated the early years of the 1970s prevented
extensive investigations into the problems related to prepositions, although there were exceptions such as Bennett 1975; Lehrer 1974; Wierzbicka 1972 and 1976; Bolinger 1977; Chafe 1968; Fillmore 1975 and 1977.
Apart from these "lonesome riders" formal aspects of language, as dealt
with in syntactic and model-theoretic analyses, were and in part still are
the sole topic of linguistic investigation. It was in the second half of the
1970s, however, that in the wake of Rosch's psychological experiments
Introduction
(cf. Rosch 1973 and 1978) an increasing number of linguists have changed
their basic attitude with respect to what constitutes the nature of linguistic
meaning.
Only a few reasons for this change in approach can be given in this
introduction. The categorical all-or-none relation between meaning and
form, which was defined by discrete categories and absolute conditions,
was only partly successful precisely because language is, to a great extent,
gradient and vague rather than absolute and precise. The predominance
of all-or-none categories in language is mirrored in the artificial distinction between semantics and pragmatics and in the exclusion of such
phenomena as metaphorical and semantic extension in general, both of
which have become central to linguistic description within the conceptual
reorientation. In addition, the usual way in which semantics is conceived
in formal theories, i. e., by reconstructing the extension ("meaning") and
the intension ("sense") of a linguistic expression in a strictly parallel
manner, has been criticized as inadequate. Thus Langacker (1987: 2)
claims that, in accordance with the conventional setting of the speaker's
environment, one and the same real situation may be associated with
different mental representations, and hence it may be assigned different
semantic values. As distinct mental images may give rise to different
expressions, the problem of how to account for these distinctions in a
semantic representation becomes especially important for the contrastive
perspective taken by MT.
Instead of being considered as contingent, as was the case before the
cognitive breakthrough, language-specific encodings are motivated by
conventionally based conceptualizations. Linguistic conventions correspond to a certain overall behaviour, which has developed in the physical,
social and linguistic context of the respective speech community. This
means that the meaning of a word relies on all knowledge resources
which may be accessed by the speakers of the respective speech community
at a specific point in time, linguistic meaning only being processed as
part of a more comprehensive cognitive routine. Hence Cognitive Grammar assumes a wide notion of context, which constrains the use of words
and their meanings. As a corollary, semantics and pragmatics have turned
out to be inseparable. This holds particularly for semantic extension to
novel word senses.
In accordance with the overall context, speakers construe the perceived
information into focus, foreground, and background. It is not only in
different languages that the overall context results in different conceptualizations and hence different expressions of the same real situation.
Also in one and the same language the same scene may be highlighted
in different ways and hence result in different expressions. That this
applies especially to the use of prepositions is illustrated abundantly in
the contributions to this volume.
2.1. Prepositions and semantic categorization

Following the tracks of Gestalt psychology, Cognitive Grammar associates semantic categories with mental gestalts. Depending on the specific
situation, these gestalts can be good or less good representations of the
respective category. The boundaries between less good, bad, and inadequate cannot be conclusively defined by necessary and sufficient conditions, as they depend on the pragmatic conditions of the respective
situation and hence may vary synchronically or be subject to change over
time. In other words, category membership is not an absolute but a
gradient phenomenon (cf. Langacker 1987: 48; Lakoff 1987).
The subjective nature of linguistic meaning does not imply total inconsistency between speakers of a language. The overall context, on which
linguistic conventions are based, continuously constrains these. The important point is that we can only interpret natural language if we explore
these overall contextual constraints and thus consider how meaning is
humanly relevant. This procedure offers the possibility of a semantic
interpretation which is psychologically real: what is meant by this is
simply that if one interprets language from a human perspective, one can
not fail to explain what it actually is that linguistic expressions refer to.
The cognitive reorientation has given birth to different paradigms
which all claim to be psychologically real. Two major paradigms are
focussed on in this volume: Cognitive Grammar, whose most prominent
exponents are Lakoff and Langacker, and the two-level approach to
semantics, as advocated by Bierwisch and Lang.
The centrality that prepositions have gained within the framework of
Cognitive Grammar relies on the fact that prepositions as spatial expressions are highly representative of the nature of linguistic meaning: conceptualization in this paradigm comprises linguistic meaning or, put the
other way round, linguistic interpretation equates with conceptualization.
The essential point is that all conceptualization, that is all mental experience, is ultimately based on physical experience gained from our physical
functioning in a spatial environment. This is what Lakoff calls preconceptual experience (cf. Lakoff 1987: 267). It is mentally encoded in image
Introduction
schemata, which function as cognitive models for all higher-order mental

constructions, that is, they give meaning even to those utterances which
deal with the most abstract and complex situations: every thought is
ultimately linked to the structure of spatial concepts which grow out of
our preconceptual structure of experience. In this view, all abstract concepts are metaphorizations and metonymies of semantically concrete
spatial predications. Abstract expressions are indirectly understood in
terms of directly meaningful, preconceptual models which are constrained
by the perception of physical relations.
2.2. The expressive force of prepositions

In conceptualizing the physical world and communicating about it we
are confronted with an immense number of spatially significant entities,
which have a specific extension in three-dimensional space according to
their dimensionality, their internal and external bounding (i. e., whether
they are bounded towards their surroundings and whether they are
internally continuous or discrete), their shape and orientation and their
mobility (i. e., whether they can change their position or not). Unbounded,
i.e. mass entities are associated with spatial properties in a restricted
sense, as they have no boundaries marking their gestalt and their absolute
extension in space and time.
It is by these spatial properties that the conceived entities constrain
the possible relations between them and hence the possible spatial predications expressed by prepositions. All understanding, communication
and experience is thereby constrained. Indeed, understanding and communication only become possible via the schematic, relatively simple
structure imposed by spatial concepts on the vast amount of information
provided by the physical and non-physical world.
2.2.1. The expression of spatial orientation
It is the characteristic capacity of the human species to move freely in
the horizontal dimension, normally in an upright position; thus human
beings can perspectivize spatial scenes from their own point of view,
thereby assigning an orientation to the objects at issue (for the following
cf. Lang 1989 and Lang this volume, as well as Langacker 1987: 122 ff.
and Vandeloise 1984: 71 ff.). The theories represented in this book agree
on the assumption that the concept of spatial orientation is determined
by the surface of the earth and by gravity as well as by our normally

upright posture. Lang therefore claims the vertical axis to be dominant
in spatial orientation (cf. Lang 1989; Lakoff Johnson 1980; Lakoff
1987). Speakers conceptualize objects in space to be oriented along their
own axes and/or in relation to another object in their surroundings. Thus
speakers assign an intrinsic orientation (cf. Vandeloise 1984: 96) to objects
which, independent of context, i.e., independent of the neighbouring
objects and independent of the dimensional grid, may receive side assignments in every position. Examples are blouse, trousers, cupboard. We have
'anthropomorphized' (cf. Vandeloise 1984: 97) these objects by categorizing them according to their intrinsic parts which we associate immanently
with them. That is, the objects are categorized with the human body and
hence with human spatial orientation as a model (cf. Wunderlich 1985:
75).
Objects may be contextually oriented either in relation to the speaker's
or hearer's orientation or in relation to some other object which is
intrinsically oriented as in the chair is in front of the table or Peter is in
front of the table (cf. Vandeloise 1984: 96 ff.). In both examples the front
may be assigned to the table both by the observer's point of view and by
the function of the located object. Lang distinguishes between contextually induced orientation, which is assigned in relation to the vertical axis
of the surrounding space and contextually induced perspectivization,
which is assigned relative to the actual or potential "observer's line of
sight" (cf. Lang 1989: 277).
When intrinsic and contextual orientation are in conflict, intrinsic
orientation prevails. The resolution of this conflict depends on the relation
by which the orientation is assigned. Vandeloise (1984: 88 and 1991)
notes that human trajectors are stronger in orienting the landmark contextually. Lang's approach to resolving this conflict is to assume a more
finegrained object categorization by what he calls "object-constitutive
primary identification" (cf. Lang 1989: 277). He distinguishes between
objects which are inherently oriented and those which are canonically
oriented. Inherently oriented objects have specified sides and can be
interpreted as being in normal position, regardless of how they are
localized in relation to the surrounding space. Examples are coin, book.
Canonically oriented objects must be in normal position in order to
fulfil their function. Objects like wardrobe, chest, tower fall into this
category. Langacker uses the term actual orientation for scenes in which
the canonical orientation is violated either by an unusual orientation of
Introduction
the speaker or by an object orientation which deviates from the norm

(cf. Langacker 1987: 123).
Finally there are objects which are neutral as to spatial orientation,
since they have no specified sides; hence they are assigned sides by
contextual cues in any position. Examples are ball, cube, star.
The speaker's position and the speaker's asymmetrical perception of
the front-back dimension comes into play with canonically oriented
entities. Thus in the sentence The table is standing in front of the cupboard
the cupboard may be given a canonical or actual orientation (the latter
may coincide with the canonical orientation if it is in line with the
speaker's/observer's point of view). The canonical orientation is independent of the speaker's point of view, but merely depends on the intrinsic
front and rear of the cupboard. This is what is generally called the
intrinsic use of the preposition in front of Actual orientation occurs in
accordance with the speaker's/observer's perspective, i.e., it arises from
his point of view, which in turn depends on his spatial or mental position.
Depending on his line of sight the speaker may assign actual sides to the
cupboard and thereby locate the table in front of the rear of the cupboard.
In this interpretation the projective preposition in front of is used deictically. Projective prepositions are generally ambiguous with respect to
deictic vs. intrinsic use. The example given with respect to this distinction
is taken from Lang (in this volume), who deals extensively with the
problems of perspectivization and orientation. To summarize, it is movable entities that speakers can perspectivize; they thereby assign actual
orientations which may deviate from the canonical ones.
Mobile or movable entities may undergo both what is called positional,
static or topological relations and directional, dynamic or projective
relations. The first type of relation expresses a position of one entity in
relation to another. The topological prepositions on, at, in for instance
relate an entity to the upper surface, the exterior side, or the interior of
a second entity. The second type of relation expresses a motion or
orientation of one entity towards another. As such, the prepositions into,
out of, from express motion or orientation into, or out of a container, or
away from an entity. In many cases there is also a topological sense
beside the projective one, as in the German equivalent of English into:
German in is used both in the topological and in the projective sense,
whereas English is lexically more specific in expressing these spatial
relations: prototypically in may be used of topological relations and into
in the projective sense; some motion verbs, as for instance go, come, walk,
run require into. Otherwise both in and into can be used in the projective
sense. Immobile entities may constitute only static relations.
Cognitive Grammar and the two level-approach agree on the assumption that these structure-imposing processes are crucial for the interpretation of and communication about the world. Their relevant hypotheses,
however, about how these processes come about, differ considerably. The
two-level approach, being object-oriented and axes-based, proceeds from
the question of how linguistic structures are determined by the orientation
of objects and by the speaker's line of sight. Cognitive Grammar departs
from the question of how linguistic expressions are determined by conceptualization and how this conceptualization is constrained by the speaker's environment, his perspective and his purposes. Cognitive Grammar
thus explains by cognitive, biological and pragmatic principles how the
speaker attributes a certain salience, relevance and typicality to the objects
and their parts which belong to a given scene, and how in accordance
with this attribution the objects participate in certain relations (for a
comparison between the two-level approach and Cognitive Grammar cf.
also Dirven Taylor 1988; Vandeloise forthcoming).
2.2.2. The expression of the foreground-background structure
Prepositions, being relational predications, profile a relation with respect
to a certain base. The base is that part of the domain which is in the
scope of the predication which is conceptually covered (cf. Langacker
1988: 58 ff. and 70 ff.). It is the function of a predication traditionally
known as head of a given construction to single out a specific aspect
of a base so that it gains special prominence. Different profilings of the
same base may be expressed by alternately valid expressions, as the
following PPs exemplify:
(1) the lorry in front of the car
(2) the taxi behind the lorry
(3) the greengrocer in front of the taxi
In these semantically contrasting expressions a common base is assumed
both with the conception of the lorry, which is of a certain type, and
with the conception of a car, which is also of a certain type, and finally
by the conceptualization of a certain spatial relation between the two
objects. A relational predication profiles the "interconnections" between
those parts of the scene which are salient in the respective discourse
situation, whereby the parts can be either "things" or some other relation.
Introduction
These interconnections represent the relative position, orientation or

motion of the parts within the domain for which the predication holds.
The relativity of some part's position or orientation is not only effected
by its entering different relations to other parts of the scene, but also by
the point of view taken by the language user: by focussing on different
parts of the same scene he moves them mentally in the foreground against
the other parts which constitute the background. This foregrounding and
backgrounding is exemplified in examples (1) to (3): whereas (1) focusses
on the location of the lorry with respect to the car, (2) turns round the
relationship and focusses on the location of the car with respect to the
lorry. (3) is metonymic in relation to (1) in that it focusses on a specific
aspect of the lorry, namely its function of transporting vegetables. It
localizes this with respect to a specific type of car, a taxi.
In each profiled relation a different part of the real scene is focussed
on as foreground; this Cognitive Grammar refers to as the trajector (tr).
This is also what is "new" in the situation of discourse. The relation
implies how the trajector is localized with respect to some presupposed
background, referred to as the landmark (lm). This establishes the "old"
or "given" part of information in the discourse situation.
This mental foregrounding and backgrounding is an old idea of Gestalt
psychology, where it became known under the terms of figure and ground,
respectively (cf., e.g., Koffka 1935; Khler 1958). However, Langacker
uses the terms trajector and landmark for the mental profilings which
are specifically constructed for expressive purposes, the trajector preferentially corresponding to linguistic realizations such as subject or verb
and the landmark corresponding to the object. Other linguists such as
Schirra and Hottenroth in this volume, and especially proponents of the
two-level approach (see Lang, Kaufmann and Pribbenow in this volume)
use the term located object (LO) instead of trajector and reference object
(RO) instead of landmark.
Although it is normally the focus of attention which determines the
partitioning of a scene into foreground and background, this partitioning
is not totally arbitrary, because certain mental organizations are highly
improbable either for gestalt reasons, or because they are unnatural or
pragmatically unusual. If, for instance, in an illustration a dark region
contrasts with its surroundings it will most probably be conceived as the
figure and not as the ground (cf. Langacker 1987: 121). The moving or
moveable object of a scene is also the one which is likely to be conceived
as foreground: in the scene in which a cat is located on a mat the cat will
usually, that is under normal pragmatic conditions, constitute the fore-
10
ground, expressed by the subject NP, while the reversed expression the
mat is under the cat sounds unusual, because the normal situation is that
the cat gets itself on the mat and not that somebody puts a mat underneath
the cat.
3. The structure and use of the lexical category preposition

The cognitive breakthrough in linguistics as initiated by Langacker,
Lakoff and other proponents of this paradigm gained its strength primarily through the conviction that the shortcomings of previous semantic
theories consisted in their adherence to absolute conditions on linguistic
meaning and their neglect of the vagueness and flexible use of semantic
categories. This cognitive reorientation was triggered by Rosch's psychological experiments on categorization (cf. Rosch 1973 and 1978), which
showed that lexical categories cannot be defined in terms of necessary
and sufficient conditions, but instead structure as sets of family resemblances. Thus the old philosophical debate about necessity, which Wittgenstein had entered with his notion of family resemblances, was rediscovered empirically.
Prepositions constitute a lexical category the elements of which are
highly polysemous. Not only do different conceptualizations of trajector
and landmark for instance with respect to bounding, dimensionality
and shape determine the respective differences between the senses of
the same relational expression, but these different polysemes are also
exploited by metaphorical and metonymic extensions into abstract target
domains so that an increasingly higher number of polysemes and increasingly abstract ones are created for the same relational expression. Dirven
in this volume illustrates this in detail. This is why prepositions can be
taken as exemplary in the study of the semantic representation and
processing of lexical units in general. The contributions to this book
approach this problem from different theoretical perspectives by touching
on the following aspects:
Which are the criteria which define the degree of conventionality,
genericity and intrinsicness by which word meaftings become established
(cf. Zelinsky-Wibbelt)?
Which are the dimensions of meaning that determine the usage patterns of prepositions (cf., e. g., Hottenroth; Dirven; Zelinsky-Wibbelt;
Lang; Kaufmann)?
Introduction
11
How finegrained is each dimension of meaning according to the

functional requirements imposed by the overall context (cf. ZelinskyWibbelt; Lang; Kaufmann; Hottenroth)?
Which are the criteria for distinguishing genuine lexical ambiguity
with different readings in the lexicon from lexical vagueness (see Hottenroth; Lang; Zelinsky-Wibbelt)?
How can we provide for all possible uses of a word? This question
relates to the problem that meaning may not be predicted in terms of
necessary and sufficient conditions, but rather by general lexical rules
and principles which can also account for unforseen word meanings (cf.
Hottenroth)?
Which are the different communicative conditions which make available different word meanings of different relevance and salience (cf.
Zelinsky-Wibbelt)?
From a translational point of view this book focusses on the mappings
between the language-specific lexicalization patterns in addition. These
may consist for instance in (cf. Givon 1978):
(a) different mappings of the same content onto different syntactic
structures (cf. Zelinsky-Wibbelt; Hawkins; Durrell Bree);
(b) mappings between different contents of source and target language
equivalent, e. g., mappings between different degrees of expressive specification (e.g., the German preposition seit which translates into the
English prepositions since and for as specialized sub-senses), or incongruence in some dimension of meaning (as with German Bus vs. bus in
English; the English equivalent still seems to keep a SURFACE sense
contrasting with German in certain situational contexts);
(c) language-specific gaps with respect to expressing some universal
semantic structure; with prepositions this problem is highly relevant:
compared to English and French German is underdetermined with respect
to expressing spatial relations by prepositions (cf. Taylor 1988 for the
relation between English and Italian; Hawkins this volume).
3.1. Categorization and context

The questions which have just been raised can only be solved by an
empirically grounded theory specifying how the content and the structure
of lexical representation is determined by the conveyance and interpretation of linguistic utterances.
12
3.1.1. Prototypes and schemata

Linguistic utterances occur in certain situational contexts which provide
the conditions for the use of a lexical unit in a specific meaning. This
statement implicitly presupposes a division of labour between two qualitatively different components of meaning:
(i) meaning that is anchored in a lexical unit;
(ii) meaning that is evoked, specialized, constrained or overridden by its
specific use in a certain context.
How this distinction is made and what status is attributed to it derives
from hypotheses about the processing of linguistic meaning and it will
crucially determine the design of a particular natural language system.
This can be observed by comparing different linguistic theories with
respect to their way of drawing the line between lexical and contextual
knowledge. In particular, different theories draw different conclusions
from this division with respect to the interpretation process. We can only
touch on this in this introduction by contrasting some basic assumptions
of Cognitive Grammar with those of the two-level approach.
The basic difference between these theories in modelling the interpretation of language derives from the following contrasting axioms: The
guiding assumption of Cognitive Grammar is the holistic and continuous
nature of all human behaviour and its underlying structure. Linguistic
behaviour is inextricably interrelated with its underlying structure, and
language in turn is continuously related to overall human behaviour and
its structure. Language is only considered as a more specific faculty
embedded in a more general psychological faculty.
In contrast to this holistic view, the two-level approach considers the
organization of human behaviour to be basically modular (cf. Lang 1987:
291). In particular, semantic and conceptual knowledge are seen as
constituting relatively autonomous though interacting systems. In a very
general sense it can be said that semantic representation and interpretation
generalizes over specializations which are only instantiated by the context
on the conceptual level.
Cognitive Grammar, in line with Prototype theory, retreats from abstracting only one common semantic representation from all possible
polysemes of a lexical unit in the lexicon, as done by the two-level
approach. Instead it pays the price of a redundant representation by
enumerating several senses in the lexicon, while additionally generalizing
over regular polysemy relations holding between word senses in the form
Introduction
13
of rules. Cognitive grammarians often emphasize that language is less

economical and systematic than it is generally claimed to be and that in
this respect a purely generative approach will not work (cf., e. g., Langacker 1987: 369 f.; Lehrer 1990). Thus Cognitive Grammar defends the
mental abstraction of a word's meaning from its instantiations (cf. Langacker 1987: 401 ff.) by categorization both in terms of prototypes and
schemata as being constitutive of the interpretation of word meaning.
Both prototypes and schemata are economical abstractions and function
as patterns of comparison in evaluating the meaning of actually used
linguistic units. 2 A prototype represents the most typical instance of a
semantic category, while a schema represents a generalization over two
or more instances. The two abstractions are closely interrelated, for
instance a schema generalizes over the prototype and its semantic extension^).
Cognitive Grammar neither distinguishes between levels of knowledge,
such as linguistic and world knowledge, nor does it single out the lexicon
or any other component as autonomous on nonarbitrary grounds (cf.,
e.g., Langacker 1987: 2,20). Instead, linguistic meaning is claimed to be
encyclopedic in nature, comprising knowledge about basic as well as
specific domains, knowledge about language use, which is related to
knowledge about social conventions and other behavioral constraints.
All knowledge structures are thus coherent with general conceptual structures. The lexicon is only seen as a more specific part of the grammar as
a whole. As such, it takes a central position within our encyclopedic
knowledge, as it relates to all types of knowledge.
A two-level model of the interpretation of meaning can be said to be
in contrast to Cognitive Grammar in all its basic axioms (for the following
cf. Bierwisch 1987 and 1989; Lang 1987 and 1989). The research paradigm
developed by Bierwisch and Lang assumes several knowledge systems
which are relatively autonomous as they have their own representations,
rules and principles. For one thing, they argue in favour of distinct
systems, because words, without being polysemous, may refer to a "whole
family of interrelated conceptual complexes" and therefore the lexical
meaning of a word should establish a "pattern generating the family of
conceptual complexes" (Lang 1989: 270). Among all systems, the conceptual system has intermodal status, as it has an integrative function
with respect to the other systems.
The semantic representation of the lexicon is constituted by the semantic system only. The conceptual system does not constitute the representation of the lexicon, but the structure of lexical entries and the
14
structure of the lexicon as part of the grammar, as there is a mutual

interaction between the conceptual and the semantic system. As the
semantic system is a proper subsystem of the grammatical system, semantic representations mediate between the grammatical and the conceptual system. A semantic representation of a lexical unit consists of
parameters which establish the potential of interpretation. This semantic
potential provides a schema for the generation of a family of concepts.
From this family a particular concept is selected depending on contextual
conditions for conceptual values which instantiate parameters of the
semantic representation. The division between semantic and conceptual
information is correlated with a division of the interpretation process
into two phases.
In contrast to this, a more holistic approach assumes that all components of knowledge interact during the whole process of interpretation.
Thus specific senses may be preferred right from the beginning. Cognitive
Grammar also assumes lexical meaning to be schematic in that one single
sense of a word is the result of a mental abstraction from all its occurrences that speakers have met in language use. However, lexical meaning
is itself conceptual and hence abstract, schematic meaning encoded in the
lexicon and its more specialized instantiations in language use cannot be
distinguished by their representation and the principles and rules working
on them. Moreover, as we have shown above, Cognitive Grammar assumes that lexical meaning is not purely schematic, but that prototypes
also establish lexical meaning abstractions.
The assumptions of these contrasting theories will lead to different
implementations of a language system. This book is an attempt to throw
some light on these differences. Many authors focus on the distinction
between lexical and contextually instantiated meaning while investigating
the process of interpreting the meaning of prepositions.
3.1.1.1. Categorization by schemata
As one of its originators, Lang presents a clear defense of the two-level
approach. In this paper he finds evidence for his approach in the necessity
to distinguish intrinsic from extrinsic uses of projective prepositions. His
essential point is that these different uses do not depend on different
semantic representations of the respective prepositions. Instead he claims
that one semantic representation is sufficient, and that the interpretation
of different concepts is the result of an interaction between the two
distinct levels representing semantic and conceptual information respectively. During this interaction the parameters which are part of the
Introduction
15
semantic representation of the preposition are assigned values by being

mapped via identification and specification onto the conceptual
representation of the reference object. On the conceptual level, spatial
objects are represented as Object Schemata, which embody information
about the object's dimensionality, shape, sidedness and orientation. The
position and location of objects in space is basically determined by what
Lang calls Primary Perceptual Space. This is a perceptually anchored
system comprising the three axes the Vertical establishing the dominant
axis, the backbone of spatial orientation, the Observer Axis and the
Horizontal.
The study continues Bierwisch and Lang's work on dimensional adjectives and is meant as an example in that the semantic properties assumed
for prepositions are claimed to have cross-categorial validity.
Adhering to the same theoretical assumptions, Kaufmann illustrates
the interpretation of prepositions with an exhaustive analysis of the
manifold uses of the German preposition durch {through). The semantic
representation generalizes over all possible uses of durch by two components of meaning: the INCLUSION component, which is the same as the
representation of the preposition in, and the PATH component, containing
the essential path parameter D (x). The inclusion component is responsible
for the interaction of the dimensionality properties of the located object
and the reference object. The information specific to the path properties
of durch is provided by the second component. The path parameter D (x)
has to be fixed by a path provided by contextual information about the
located object x. The actual course of the path is then determined by the
further interaction of the conceptual information associated with the
located object and the reference object. At least three types of information
are encoded by the concepts associated with nouns denoting objects:
gestalt properties, functional properties (at least as far as artifacts are
concerned) and structural properties of the object. This information can
be used to fix the parameter D (x).
Pribbenow develops a two-level model of the computational interpretation of localization expressions. This has been implemented in the
LILOG prototype LEU 2, a text understanding system with the ability
to answer questions. Localization expressions are defined by their function of predicating the static or dynamic localization of entities, which is
typically designated by prepositions. The paper focusses on the conceptual
analysis of localization expressions, exemplified by prepositions, and on
the influence of the textual and situational context. The adequate processing of knowledge is achieved by using picture-like representations,
16
referred to as depictions. The depictions are combined with a propositional formalism, which interprets general regularities between spatial
concepts and non-spatial, e. g. functional aspects of meaning.
The representation of the conceptual meaning of localization expressions is assessed by areas functioning as search domains with respect to
the entities to be located. Each search domain consists of a primary area
describing the basic meaning of the localization expression and priorizations which take into account expectations about the typical localization
of the entity to be located. The search domain is the result of the
depictorial interpretation working on the concept combination, which
has been computed by the propositional rules.
3.1.1.2. Categorization by prototypes
The semantic model proposed by Hottenroth is a hybrid model in the
sense that, on the one hand, she also starts from the assumption of a
two-level model of semantics, but on the other hand, she does not assume
that the semantic representation of the preposition has to be an abstraction from all possible contextual interpretations, but only from a certain
range of "prototypical" uses. In a detailed analysis of the spatial uses of
the French preposition dans (in), Hottenroth shows that the adoption of
such an "overspecified" semantic representation turns out to be quite a
natural and psychologically plausible assumption. The analysis of the
process of interpretation gains its plausibility by taking into account the
part played by certain general cognitive principles known from the domain of object conceptualization and categorization, as well as certain
systematic manipulations of object concepts. The paper shows how these
principles determine the interpretation of prepositions dependent on their
different possible reference objects. Within this model the author makes
explicit what "similarity" or "tolerance" means in linguistic categorization.
In analyzing the semantics of the Dutch preposition in, Cuyckens
proceeds from the view that meaning is encyclopedic in nature. Thus he
describes the differing senses of in as a family resemblance structure and
explains the fuzziness at the borderline of the senses as a natural consequence of man's functioning in his social environment. On the one hand
he assumes that the abstract relation COINCIDENCE generalizes over the
different relations between trajector and landmark expressed by the
preposition in. On the other hand he assumes that Hawkins' spatial
configuration MEDIUM subsumes differing, but related, configurations of
the landmark. Cuyckens analyzes these subconfigurations in terms of
Introduction
17
internal and external bounding and in terms of the dimensionality of the

landmark. He concludes that porous, three-dimensional bounded mediums figure as prototypical in the family of senses of in. The sense of in
which refers to a part-whole relationship is outside this family resemblance
structure, but is still related to the chain of meaning via the configuration
of the landmark.
The path uses of in copy the family resemblance structure analyzed for
the non-path uses of in and the two are related by transformations as
well as by conceptual intersection.
In dealing with image understanding Schirra explores and models an
experientialist relation between natural language and perception. The
topic of this paper is how the experientialist relation is processed by the
SOCCER system, the aim of which is the interpretation and German
verbalization of visual information provided by soccer games. This is
achieved by visually perceiving locations and movements, interpreting
conventions, assuming intentions and plans and finally selecting events
for adequate verbalisation. All this is achieved in relation to the participants of the game.
By translating visual percepts into prototypically structured spatial
concepts defined by a conjunction of graded functions of essential parameters, SOCCER represents conceptual vagueness: each spatial preposition is associated with a core meaning very similar to Herskovits' ideal
meaning; from this core meaning actual meanings are derived by algorithms, resulting in a classification function, which associates SOCCER's
percepts to a graded Spatial Concept corresponding to the usage of a
spatial preposition. Based on a function over applicability degrees in
generation, linguistic hedges are used to express good or less good
instances of such a graded concept. Schirra shows how in reverse these
structures can be used for the construction of mental images during the
interpretation of PPs.
Aurnague and Vieu present a text analysis system in which inferences
are drawn in analogy to human reasoning. Although their semantics is
modularized into three qualitatively different levels, the authors rely
heavily on Herskovits and Yandeloise. At the geometrical level, purely
objective topological and projective properties are represented. This
knowledge is used for the interpretation of the orientation and structure
of entities as well as some common sense spatial relations at the functional
level. The internal structure of entities is described in terms of several
part-whole relations and their ordering relations, which the authors
explain by Link's lattice theoretical analysis of entities. By these relations
18
Aurnague and Vieu can distinguish functionally different senses of the

preposition dans (in) and sur (on). The representation of geometrical and
functional relations is done in first-order predicate logic. While the
knowledge represented at these two levels is indefeasible, this does not
hold for the third level, which represents pragmatic principles. Relying
heavily on the function of objects, these pragmatic principles filter out
relations wrongly inferred and add information by using knowledge about
the typical use of objects, their relevance in relation to other objects and
the quantity of objects. By taking into account these additional conceptual
relations, Aurnague and Vieu succeed in inferencing relations which could
not be accounted for in purely geometrical and functional terms. As there
may always be highly marked situations in which the typicality and
relevance normally attributed to a sense may be defeated, the pragmatic
knowledge may only be manipulated by non-monotonic logic; this, however, was still being developed at the time of writing.
3.1.2. Semantic extension of lexical meanings
The phenomenon of semantic extension has currently undergone a revival
not only in Computational Linguistics with respect to the implementation
of a lexicon, but also in Linguistics in general. It is being discussed in
regard to the options of an economical vs. a more redundant lexical
representation. The economical representation would generalize over different senses of a lexical unit by representing only the most abstract
meaning in the lexicon, and by representing the polysemous structures
by rules which generate the polysemous senses. The latter is obviously
very helpful for the interpretation of novel senses which are not foreseen
in the lexicon. Still, it may be asked whether such a redundancy-free
representation is sufficient. Many linguists, often those who proceed from
an empirically grounded work, refute this position (cf., e. g., Lehrer 1991)
with the argument that language although regular to a high degree
is not completely regular. This is especially true of Cognitive Grammar.
Cognitive Grammarians generally agree that meaning extension by metaphor and metonymy constitutes productive processes operating on the
lexicon (cf. Lipka 1988: 357). Yet, it contradicts the philosophy of Cognitive Grammar to set up categorical rules for patterns of meaning
extension which can predict each possible meaning extension with complete certainty (cf. Taylor 1989). Rather, these predictions should hold
for preferred patterns of meaning extension. Dirven gives a good example
of this in this volume. He illustrates chains of meaning resulting from
Introduction
19
meaning extension. Meaning extension proceeds from a prior meaning

which is projected onto increasingly abstract senses in various cognitive
domains. It is the intention of this article to point out the semantic
regularities holding between different cognitive domains such as time,
state, area, manner or means, circumstance, cause or reason on the one
hand. On the other hand each domain is investigated with respect to the
relations existing between the concepts expressed by different prepositions. The result is that there are gaps in meaning extension in some
abstract domain, because of the underlying spatial sense relations of the
respective preposition which have conventionalized a particular structure.
Taylor's article allows for a related conclusion: he analyzes polysemy
relations for an extensive set of English prepositions and sets up patterns
which, though they are realized by a large class of prepositions to a high
degree, do not hold without exception. In fact, he discovers many meaning
extensions which pertain to one preposition only.
He classifies prepositions with respect to the extensions they can undergo on the basis of the profiled relation. Both simplex (on, under, at)
and multiplex prepositions (around, along, through) can be categorized
with respect to the characteristics of their trajectors on the one hand and
their landmarks on the other. Multiplex prepositions are more productive
and variegated with respect to the semantic extensions they can undergo.
The paper closes with a catalogue of knowledge types, which Taylor
assumes as necessary for the selection of the intended lexical meaning:
(1) world-knowledge;
(2) knowledge about the valency of other constituents in the sentence,
especially the verb;
(3) knowledge about the sentential aspect;
(4) knowledge about the informativity of the utterance.
In the framework of X-bar theory Rauh analyzes English prepositions
by the relational properties which they exhibit in the grammatical constructions in which they participate. By analyzing the different types of
uses of prepositions according to their relational properties, she points
out the heterogeneous functions of prepositions. First, lexical prepositions
are distinguished from non-lexical prepositions by the syntactic and
semantic properties which they have as heads of constructions and by
their co-occurrence restrictions, which they impose on other constituents
within their maximal projection.
Non-lexical prepositions are analyzed as constituting different intermediate stages in the bi-directional processes of lexicalization and gram-
20
maticalization originating in a marked metaphorical use of a form and

ending in a lexical variant. Case prepositions are an instance of grammaticalization of originally autonomous lexical forms. Prepositions in
fixed phrases are grammatical functions taking part in "adjectivizing" or
"adverbializing" lexicalization processes. These multi-word units are associated with lexical entries of their own as well as with lexical rules
which describe their syntactic and semantic regularities.
Existing metaphorical extensions of the spatial meanings of prepositions may be observed in different stages. Transfer into the temporal
domain results in a reanalysis of the lexical properties in accordance with
the temporal domain and in lexical variants with individual lexical entries.
Other metaphorical extensions such as scalar and sequential uses of
basically spatial prepositions constitute lexicalized conceptual variants
which are the result of a reinterpretation by imposing the spatial structure
on the target domain.
Rauh's analysis concludes that an account of grammatical structures
only in terms of the cognitive principles imposed on them is insufficient,
as the same metaphorical processes which are determined by the cognitive
principles, result in different grammaticalizations. Therefore it is rather
the grammaticalization process itself which restricts grammar.

Cognitive Grammar offers particular benefits for machine translation as
it allows for both universal and language-specific semantic structures to
be analyzed and related. It assumes that there is universality of meaning
on the one hand resulting from a cognitive endowment which is common
to all language users of the world. On the basis of cross-cultural differences, however, which provide a different environment to the language
user, this common cognitive endowment is exploited in different ways:
different distributions of universal dispositions are mirrored in different
patterns of meaning in different languages.
Hawkins studies the interaction of both universal and language-specific
meaning from a contrastive point of view: he analyses the spatial adpositions of the two languages English and Japanese, the cultural environments of which are maximally distinct. Not surprisingly, he finds that
the two languages make different use of certain semantic resources, while
being compatible with respect to others. He obtains this result by assum-
Introduction
21
ing profilable structures which correspond to the common cognitive

endowment and which are not developed with respect to specific cognitive
domains. These innate structures comprise the basic spatial relations of
SAME and DIFFERENT and the two parameters of relation and configuration.
By contrasting this universal structural potential against the actually
profiled structures of the specific languages Hawkins brings out their
differences and correspondences in configurational and relational information.
Zelinsky-Wibbelt confronts the problem of cross-language variations
and correspondences with the translation of prepositions. She claims that
in natural language processing and in machine translation in particular
a semantic representation needs to be conceptually constrained in order
to interpret and generate different usage patterns in different languages.
By proceeding from the assumption that the conceptualization of mental
space is perceptually driven she illustrates how pragmatic principles
concerning the function of entities determine the salience, relevance and
typicality of the entities constituting the conceived situation. These principles determine the process of schematization, in which the spatial
relation expressed by the preposition follows from the spatial properties
of the related entities. These in turn are derived from prototypical and
schematic categories encoded in the lexicon, all in accordance with the
utterance situation and the domain of discourse. Zelinsky-Wibbelt assumes that Hawkins' PLACE relations of COINCIDENCE and SEPARATION
constitute the "tertium comparationis" which is necessary for translating
prepositions between different languages. These relations are abstracted
from the profiled relation of the source language and handed over to the
target language. The target language uses this transferred information
together with its own specific lexical and compositional rules for the
generation of the correct preposition.
Zelinsky-Wibbelt formalizes in a model how the different knowledge
resources are dependent on each other in mapping prepositions from
German into English in a machine translation system.
Not committing themselves to a specific theory, Durrell and Bree
explain the use of temporal conjunctions and prepositions by selection
procedures which they represent in selection trees. They proceed from
Bree's earlier proposals for English and Dutch and develop a framework
for German. The selection trees determine the decision process for the
correct use of a temporal conjunction or preposition by temporally
relating the time of the state or event expressed by the matrix and
subordinated clause and by the time of discourse. The cross-language
22
comparison as to the expressive distinctions by temporal conjunctions

and prepositions systematically contrasts the differences and explains
them as the result of
different conditions being focussed on in the decision processes in
different languages;
different conceptualizations of the temporal extension referred to by
the nominal governed by the preposition.
The high number of German prepositions which are used both in a
durative and a non-durative sense has led to two selection trees in
German. The validity of the trees might be tested by adapting them to
an NLP system, which would be easy to do as the format of the trees is
very near to computer interpretable rules.
Letting the reader now go in medias res, we hope that this introduction
has shed some light on the area of research presented in this book. In
particular, seeing the area of research in terms of mental space, we hope
that it has become clear how different semantic viewpoints or perspectivizations of the topic of research can lead to different theoretical conceptualizations. The editor apologizes for a presumably biased presentation,
due to the fact that she herself subscribes to the theory of Cognitive
Grammar.
Notes
1. I would like to thank all colleagues who have discussed these introductory notes with
me. I am especially indebted to Ewald Lang, Rene Dirven, John Taylor, Hubert Cuyckens,
Claude Vandeloise, Erich Steiner and Michael Grabski for their time and advice which
is reflected in this introduction. Thanks are also due to John Taylor and J. A. Charlton
for giving the finishing touches to my English. Of course I am responsible for all
remaining errors.
2. Langacker uses the term usage event for this comparison.
References
Bennett, David C.
1975
Spatial and temporal uses of English prepositions An essay in stratificational
semantics. London: Longman.
Bierwisch, Manfred
1987
"Semantik der Graduierung", in: Manfred Bierwisch Ewald Lang (eds.),
91-286.
Introduction
1989
23
"The semantics of gradation", in: Manfred Bierwisch Ewald Lang (eds.),

71-261.
Bierwisch, Manfred Ewald Lang (eds.)
1987
Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven.
Berlin:
Akademie-Verlag.
1989
Dimensional adjectives. Grammatical structure and conceptual interpretation.
Berlin: Springer.
Bolinger, Dwight
1977
Meaning and form. New York: Longman.
Chafe, Wallace L.
1968
"Idiomaticity as an anomaly in the Chomskyan paradigm", Foundations of
Language 4: 109-127.
Cole, Peter Jerrold M. Sadock (eds.)
1977
Syntax and semantics. Volume 8: Grammatical relations. New York: Academic
Press.
Dirven, Rene John Taylor
1988
"The conceptualization of vertical space in English: The case of taW\ in:
Brygida Rudzka-Ostyn (ed.), 379^102.
Fillmore, Charles J.
1975
Santa Cruz lectures on deixis. Bloomington: Indiana University Linguistics
Club.
1977
"The case for case reopened", in: Peter Cole Jerrold M. Sadock (eds.),
59-81.
Givon, Talmy
1978
"Universal grammar, lexical structure and translatability", in: Franz
Gnthner Monica Gnthner-Reutter (eds.), 235-272.
Gnthncr, Franz Monica Gnthner-Reutter (eds.)
1978
Meaning and translation. Philosophical and linguistic approaches. London:
Duckworth.
Hllen, Werner Rainer Schulze (eds.).
1988
Understanding the lexicon. Meaning, sense and world knowledge in lexical
semantics. Tbingen: Niemeyer.
Khler, W.
1958
Dynamische Zusammenhnge in der Psychologie. Bern: Huber.
Koffka, K.
1935
Principles of Gestalt Psychology. London: Harcourt, Brace & World.
Lakoff, George
1987
Women, fire, and dangerous things. What categories reveal about the mind.
Chicago/London: University of Chicago Press.
Lakoff, George Mark Johnson
1980
Metaphors we live by. Chicago: Chicago University Press.
Lang, Ewald
1989
"The semantics of dimensional designation of spatial objects", in: Manfred
Bierwisch Ewald Lang (eds.), 263-417.
Langacker, Ronald W.
1987
Foundations of cognitive grammar, Vol. 1. Theoretical prerequisites. Stanford:
Stanford University Press.
1988
"A view of linguistic semantics", in: Brygida Rudzka-Ostyn (ed.), 49-90.
24
Lehrer, Adrianne
1974
Semantic fields and lexical structure. Amsterdam: North Holland.
1991
"Polysemy, conventionality and the structure of the lexicon", Cognitive Linguistics 1-2: 207-246.
Lipka, Leonhard
1988
"A rose is a rose is a rose: On simple and dual categorization in natural
languages", in: Werner Hllen Rainer Schulze (eds.), 355-366.
Moore, Timothy . (ed.)
1973
Cognitive development and the acquisition of language. New York: Academic
Press.
Rosch, Eleanor
1973
"On the internal structure of perceptual and semantic categories", in: Timothy
E. Moore (ed.), 111-144.
1978
"Principles of categorization", in: Eleanor Rosch Barbara B. Lloyd (eds.),
27-47.
Rosch, Eleanor Barbara B. Lloyd (eds.)
1978
Cognition and categorization. Hillsdale, N. J.: Erlbaum.
Rudzka-Ostyn, Brygida (ed.)
1988
Topics in cognitive linguistics. Amsterdam/Philadelphia: John Benjamins Publishing Company.
Schweizer, Harro (ed.).
1985
Sprache und Raum. Ein Arbeitsbuch fr das Lehren von Forschung. Stuttgart:
Metzlersche Verlagsbuchhandlung.
Taylor, John
1988
"Contrasting prepositional categories: English and Italian", in: Brygida
Rudzka-Ostyn (ed.), 299-326.
1989
Linguistic categorization: Prototypes in linguistic theory. Oxford: Clarendon
Press.
Vandeloise, Claude
1984
Description of space in French. Doctoral Dissertation, University of California,
San Diego: L . A . U . D . T . 1985.
1991
Spatial prepositions: A case study in French. Chicago/London: The University
of Chicago Press.
in press
"Methodology and analyses of the preposition", in: Leuvense Bijdragen.
Wierzbicka, Anna
1972
Semantic primitives. Frankfurt: Athenum.
Wittgenstein, Ludwig
1953
Philosophical investigations. Oxford: Blackwell.
Wunderlich, Dieter
1985
"Raum, Zeit und das Lexikon", in: Schweizer, Harro (ed.), 66-89.
1. Lexicalization patterns of prepositions
The Dutch spatial preposition "in":

A cognitive-semantic analysis
Hubert
Cuyckens
1. Setting the scene

This paper provides a lexical-semantic analysis of the spatial senses of
the Dutch preposition in, or, in short, of the spatial preposition (SpP) in,
as it is exemplified in (1):
(1)
a. De vaas Staat in de kast.

'The vase is in the cupboard.'
b. Hij woont in onze buurt.
'He lives in our neighborhood.'
c. Teken een lijn in die cirkel.
'Draw a line in that circle.'
d. De knik in de curve is heel significant.
'The dip in the graph is very significant.'
Before embarking on the analysis proper of in, we will (a) look at the
nature of spatial prepositions in general, and (b) briefly present the lexicalsemantic framework within which this study is situated.
1.1. Spatial prepositions defined

Spatial prepositions indicate the spatial relation between two arguments
and y, i.e., how and y relate to each other in space.1 (In cognitive
linguistic studies, it has become customary to utilize the terms trajector
(TR) and landmark (LM) 2 for the arguments and y, respectively.) More
specifically, in a large number of cases, spatial prepositions describe the
place/location of (in other words, they assign to a particular place)
by using the argument y as reference, or still, they serve to locate with
respect to y, or rather, the place of y. Examples are:
(2)
a. the books on the table

b. He washed his car in the garage.
28
Hubert Cuyckens
On in (2 a) locates the books (argument x) with respect to the table

(argument y): it assigns to a place or location which coincides with the
the place taken up by top of the table (a place which is assumed to be
known). The English SpP in (henceforth inE) in (2 b) stipulates that the
location of the event expressed in the sentence He washed his car coincides
with the interior of the garage.
In a number of cases, though, SpPs serve not to locate the entire entity
with respect to y, but an element in its semantic structure. Consider
(3):
(3)
a. The children were drawing on the wall,

b. the sign to Brussels
In (3 a) it is only the result of the children's drawing which is located

with respect to the wall. In (3 b) not the sign is assigned a particular
location, but the direction this sign adopts.
1.2. Classical vs. cognitive lexical semantics

Irrespective of the lexical-semantic framework we adopt, we take lexical
meaning or lexical-semantic information to be conceptual in nature. Word
meanings or lexical concepts are mental entities that link, or mediate
between, words and their (extralinguistic) denotata. More precisely, a
lexical concept X (labeled by the word x) establishes a class/category
"X" of denotata that fit the conditions (or conceptual information) in
X. 3
Interesting proposals for the semantic analysis of lexical items have
recently been put forward within cognitive linguistics. Indeed, cognitive
lexical semantics has become a widely accepted alternative to the socalled classical 4 view of word meaning.
On the whole, these two lexical-semantic approaches differ in three
important respects: 5
(a) Classical lexical semantics advocates the existence of an autonomously linguistic level of word meaning and, concomitantly, the separation of linguistic-semantic and encyclopedic information in the conceptual
content of a word. Cognitive lexical semantics, in contrast, suggests that
all conceptual information associated with a lexical item is broadly
encyclopedic in that it is part of, and needs to be understood against the
background of, broader cognitive structures. On this view, word meaning
The Dutch spatial preposition in
29
is not determined by the language system itself, but reflects how people
interact with, perceive, and conceptualize the world.
(b) With respect to the structure of word meaning, the classical approach holds that (autonomous) lexical concepts are well-delineated entities whose definitions are couched in terms of an invariable set of
necessary and sufficient features applicable to all the instances in that
concept. From this definitional perspective, then, all instances are completely identical. 6 Now, on the cognitive view, instances of a concept may
be linked not because they all share the same features, but because they
share different sets of features with each other; in other words, because
they are similar to each other in different respects (or along different
dimensions), very much like the members of one family. As such, what
links the various instances of a lexical concept is a family resemblance
relationship.
The structure of lexical concepts must capture this family resemblance
relationship between instances. A simple list of (encyclopedic) features is
an inadequate representation of a family resemblance concept (even when
allowance is made for the fact that none of these features need be
necessary and/or sufficient), because it does not sufficiently convey that
different (yet, mutually related) featural configurations may apply to
different subsets of instances of the concept. Therefore, as Geeraerts
(1989) suggests, family resemblance concepts can be represented as a
cluster of overlapping featural configurations. 7 Instead of representing
lexical semantic structure in terms of featural configurations, one might
also resort to imaginal representations (depictions of the physical world),
with overlapping imaginal representations corresponding to overlapping
featural configurations (cf. Brugman 1981; Lakoff 1987).
(c) While classical lexical concepts are made up of an invariable set of
criterial features, all of which have an equal status in defining the concept,
some featural information in cognitive concepts may be more salient or
prototypical than other (extensionally, some concept instances are more
prototypical, salient, or representative than others) (cf. Rosch 1975, 1978;
Rosch - Mervis 1975).
It has been amply shown elsewhere that the classical account of word
meaning runs into a number of serious problems and that the cognitive
approach constitutes a far more satisfactory framework for lexical-semantic analysis (Haiman 1980; Lakoff 1982, 1987; Geeraerts 1985: 82148, 1986: 112-115, 1988, 1989; Cuyckens 1991). It goes without saying,
then, that also the lexical item in should be analyzed within this cognitive
framework. While it is beyond the scope of this paper to dwell at length
30
Hubert Cuyckens
on the merits of the cognitive approach, it might be instructive, given

that in is an SpP, to look at some evidence from earlier studies of SpPsemantics against classical and in favour of cognitive lexical semantics.
(a) Also in the semantic study of SpPs the autonomy requirement
(typical of classical lexical semantics) is no longer tenable.
(i) A significant number of (allegedly) non-encyclopedic SpP-deflnitions
exploit the properties associated with either one of the SpP's arguments
(usually of argument y). I doubt, though, whether these properties can
be considered to be properly semantic (i. e., non-encyclopedic) ones, when
or y are not part of a prepositional construction. InE and on, for
instance, are often said to describe a COINCIDENCE relation between
and a 3DIM container or 2DIM surface y, respectively.
(4)
a. the keys in the cupboard

b. the statue on the Marketplace
I have a strong hunch that advocates of the classical view would have to
regard these physical and functional attributes of y (cupboard: potential
container, 8 3DIM; Marketplace: 2DIM, surface) as encylopedic. Now,
advocates of the classical view could still argue that otherwise encyclopedic information associated with a lexical item becomes non-encyclopedic when this item is the argument of an SpP. In my mind, though,
one cannot seriously maintain that the same conceptual information is
now encyclopedic then semantic. Similar examples can be found in (5):
(5)
a. The child crossed the street in front of my car.

b. the trees along the road
In (5 a)'s most likely interpretation, (i. e., the child crossing the street)
is located at some point on the axis defined by the inherent front of the
car. Now, the fact that a car has an inherent front would, on the classical
view, fall outside the strictly semantic information associated with car.9
In (5 b) along exploits an encyclopedic attribute of x, viz., its significant
extension in one dimension.
(ii) The encyclopedic information associated with y may also play a
crucial role in resolving the ambiguity of prepositional expressions. Consider (6):
(6)
a. a ring around the bathtub

b. a ring around the collar
(example taken from Hawkins 1985: 287)
31
Strictly speaking, the prepositional constructions in (6) involving around

are ambiguous: may describe a circular path (i) which coincides with y
(e. g., in (6 a) the visible residue of soap or dirt that is left when the water
in the tub has drained and in (6 b) the residue of dirt when a shirt has
been worn for too long) or (ii) which is separate from y (e. g., when the
householder has drawn a circle around the bathtub). Now, the first
reading is the preferred one because it is part of our practical (or
encyclopedic) knowledge that rings of dirt/soap residue are a typical
feature of bathtubs.
(iii) I pointed out above that on the cognitive view, all conceptual
information associated with a lexical item is broadly encyclopedic in that
it is part of, and needs to be understood against the background of,
broader cognitive structures. In Langacker's terminology, the semantic
structure of a word necessarily takes the form of a profile or figure which
is characterized relative to a base or ground.
With respect to SpPs now, it seems quite obvious that physical space
has a prominant role in the base of or the background against which
SpPs must be understood. How do we understand physical space? First
of all, there is the Newtonian conception of space "as an infinite, continuous, stationary, three-dimensional box [that] enables the speaker to label
locations by their coordinate values ... (given a point of origin)" (Miller
Johnson-Laird 1976: 380). However, ordinary languages deal with
relativistic space, space in which the location of objects is determined
relative to other objects. Our three-dimensional conception of space can
be attributed to our visual appreciation of depth and of the horizontal
and vertical relations between objects (cf. Miller Johnson-Laird 1976:
57-76). Now, neither actual physical space nor our conception thereof
has an inherent orientation. Oriented physical space requires laying out
a three-dimensional coordinate system relative to an origin.
Not all SpPs talk about space relative to an origin and coordinate axes
(e.g., inE, on, at). Yet, in front of, behind, to the left of, to the right of,
above, and under crucially involve the imposition of a three-dimensional
matrix on physical space. Furthermore, in front of, behind, to the left of,
and to the right of can be interpreted relative to a speaker's egocentric
origin (deictic use) or relative to coordinate axes derived from the intrinsic
parts of y itself (intrinsic use).
Another element in the base of SpPs is the observer. In He lives across
the bridge, he ( = x) must be located at the end of a mental path that
initiates at the location of an observer (on this end of the bridge) and
that terminates at the other end of the bridge. Also past in He lives past
32
Hubert Cuyckens
the post office and the deictic uses of in front of, behind, to the left of,
and to the right of involve this observer.
(b) The second "classical" requirement, viz., that lexical items be
characterized in terms of a set of necessary and sufficient conditions, also
presents difficulties for the semantic analysis of SpPs.
(i) For quite a few SpPs, a description in terms of criterial features can
simply not be found. An amply studied example is over, which involves
at least three senses: "above and across", "above", and "covering" (cf.
Brugman 1981; Lakoff 1987).
(7)
a. We flew over the Atlantic.

b. the picture over the door
c. the tablecloth over the table
Also the Dutch SpP op, a number of whose uses are exemplified in (8),
is not amenable to a necessary and sufficient description.
(8)
a. Het pakje op de tafel

'the package on the table'
b. de mooie fresco's op de zoldering
'the frescos on the ceiling'
c. het ongeval op het kruispunt
'the car accident at the intersection'
d. Hij is op zijn kantoor.
'He is at his office.'
Given, on the one hand, the highly polysemous nature of most SpPs and
the difficulty to find a single set of criterial features on the other, classical
lexical semantics would be forced to treat the different senses of SpPs as
homonyms. I find this a position hard to maintain.
(ii) The necessary and sufficient descriptions that are actually proposed
are not adequate. Let us have a look at inE, for which Bennett (1975)
and Cooper (1968) propose an analysis in terms of the notion 'interior'.
This description is neither necessary nor sufficient. First, the feature
"interior" is not involved in all uses of inE, as the examples in (9) show:
(9)
a. the oasis in the desert

b. He got a kick in the stomach.
Moreover, not all spatial configurations that can be classified as an

interior give rise to inE. Consider the following situations taken from
Vandeloise (1986: 232-233):
33
Figure 1. The smoke in the cheese-cover
In Figure 1, cheese-cover features an interior, viz. the 3DIM, non-porous

space bounded by its outer edges and the table. The same spatial configuration occurs in Figure 2; still, under is the preferred SpP.
(iii) The classical description of SpPs creates the impression that SpPs
are clear-cut, well-delineated concepts. However, a fair number of SpPconcepts are fuzzy:
(10)
a. The plane flies over the mountain,

b. Y is behind X.
When the path of the plane in (10 a) is directly above the peak of the
mountain, the use of over is sanctioned. Suppose now that the path of
the plane is rather to one side of the mountain (cf. Figure 3). If it is
reasonably near the top, over is still appropriate, but there is no clear
demarcation line indicating where the use of over is still sanctioned and
where it no longer is. Similarly in (10 b), if y is not directly behind x, it
is unclear from exactly what point onwards the spatial relation between
and y must be lexicalized by means of to the right of or to the left of
instead of behind (Figure 4).
Summing up, from the preceding remarks, it must have become clear
that it certainly seems justified to adopt a cognitive-semantic framework
for the analysis of the Dutch SpP in.
1=1
Figure 2. The pear under the cheese-cover
34
Hubert Cuyckens
in f r o n t
to
the
left
of
Figure 4. Y is behind X
of
to
t h e right
>
of
35
2. The analysis of "in": An outline

In the rest of this paper, I will be concerned with the cognitive-semantic
analysis of the spatial relation expressed by the preposition in. More in
particular, I will concentrate on the lexical-semantic structure of in and
on the isolation of a prototype.
Here are a number of representative examples:
(11)
a. De vaas Staat in de kast.

'The vase is in the cupboard.'
b. Let op voor de nagels in de plank.
'Watch out for the nails in that board.'
c. Hij woont in onze buurt.
'He lives in our neighborhood.'
d. Teken een lijn in die cirkel.
e. Hij woont in Duitstand.
'He lives in Germany.'
f. Hij zette de champagne in de ijskast.
'He put the champagne in the fridge.'
g. De knik in de curve is heel significant.
'The dip in the graph is very significant.'
It is obvious from these examples that in expresses a relation of COINCIDENCE between the arguments and y (i. e., y constitutes the place
where is located). Sentences (11 a-f) differ from (11 g) in that the latter,
next to expressing COINCIDENCE, also involves a part-whole relation
between and y. Furthermore, it can be observed that in lexicalizes static
spatial relations (11 a-e, g) as well as relations denoting a path (in the
latter case, COINCIDENCE holds only between the terminus of the path
traversed by and y).
The analysis of in will be organized as follows. The first, and largest,
part (Section 3) will address the static or non-path uses10 of in (cf. (11 a e, g)). Within this section, the following issues will be dealt with:
1. I will first describe in expressing non-part-whole (NPW) relations. In
this respect,
(a) I will shortly look at existing analyses of the equivalents of the Dutch
in in cognate languages. From this survey it should become clear that (i)
the two main elements in the semantics of in are the relation COINCIDENCE and the spatial configuration "medium" associated with the
36
Hubert Cuyckens
entity y; and that (ii) this spatial configuration should not be described
by way of necessary and sufficient conditions, but in terms of a family
resemblance structure (i. e., in terms of different, yet interrelated subsets
of features). (Section 3.1.1.)
(b) I will examine in detail the various types of medium, or, as the
meaning of in crucially hinges on the notion "medium", the different
meanings of (NPW-)/w. (Section 3.1.2.)
2. I will then briefly go into the use of in expressing part-whole (PW)
relationships.
3. I will demonstrate how the different types of NPW-senses and the PWsense of in interrelate, that is, what combines them into a family resemblance structure.
4. Although the family resemblance concept "medium" is the backbone
of the semantics of in, a number of additional semantic aspects of in
(e.g., fuzziness) will also be looked at.
5. I will isolate the prototypical instances of the static SpP in.
In a second part (Section 4), I will briefly examine the uses of in expressing
a path 1 1 and investigate how they complement the family resemblance
structure arrived at for the static uses of in.
3. The analysis, of "in" as a static SpP

3.1.
In this first part, I will largely concern myself with the analysis of in
lexicalizing the spatial relation between two discrete arguments and y.
3.1.1.
Although we described (11 a-e) earlier as clearly expressing a COINCID E N C E relation, it is equally obvious from these examples that a semantic characterization of in in terms of C O I N C I D E N C E alone is not
sufficient: in (12) the arguments each enter into a COINCIDENCE
relation with the arguments y, but SpPs other than in are employed.
(12)
a. De boeken liggen op de tafel.

'The books are on the table.'
37
b. Er hangt een briefje aan de deur.

'There is a note on the door.'
In (12 a) coincides with an argument y exhibiting a spatial configuration
different from the one normally attributed to in ; in (12 b) the relational
content of the SpP involves ATTACHMENT rather than mere COINCIDENCE. 12
What, then, constitutes an adequate semantic description of in (and,
concomitantly, what distinguishes in from the other SpPs involving COINCIDENCE)? Before presenting my own proposal, I would like to look
at a number of possible characterizations of the Dutch SpP in based on
existing descriptions of its counterparts in other, cognate languages, viz.,
in in English and dans in French, and examine whether these analyses
provide an adequate account of the semantics of in (in Dutch). 13 (So,
when we refer below to "Bennett's analysis of in", for instance, we regard
that as short for "the analysis of the Dutch SpP in based on Bennett's
findings".)
3.1.1.1.
One obvious way to come to grips with the semantics of in is to describe
it, along with the SpP op 'on, at', in terms of a rather abstract COINCIDENCE relation between and y. What seems to distinguish in from
these other prepositions is the spatial (in particular, the dimensional)
characteristics of the entity y. (As such, the dimensional properties of
are considered irrelevant.) Clark (1973) and Quirk and Greenbaum (1973)
describe inE in this fashion: they claim that inE lexicalizes a COINCIDENCE 1 4 relation between and (a) a 3DIM entity (Clark) or (b)
a 3DIM volume or a 2DIM area (Quirk Greenbaum).
Neither of these descriptions provides an adequate basis for the semantics of in.
(a) Clark's analysis is probably the most rudimentary. First, his requirement that y be a 3DIM entity clearly does not account for the use
of in with 2DIM and 1DIM entities:
(13)
a. In de woestijn vind je heel wat slangen. (2DIM)

'There are quite a few snakes in the desert.'
b. Pas op voor die bocht in de weg. (1DIM)
'Watch out for that curve in the road.'
Quirk and Greenbaum score considerably better in this respect: although

they do not explain the use of in with 1DIM entities either, their char-
38
Hubert Cuyckens
acterization of inE in terms of a COINCIDENCE relation with a 3DIM

volume or a 2DIM area offers a sounder basis for the analysis of in. In
fact, it covers the large majority of spatial relations lexicalized by in,
because, as we will see further, the use of in with 1DIM entities is restricted
to a specific type of part-whole relations.
(b) Neither Clark's nor Quirk and Greenbaum's characterizations make
it sufficiently clear that, when the entity y is seen in its full-fledged
dimensionality,15 often only a particular portion of y, and not the entire
entity as such, is involved in (or enters into) the COINCIDENCE relation
with expressed by in. In other words, the potential range of locations
that an SpP such as in assigns to constitutes only a portion of y. In
this respect, consider (14).
(14)
a. Het geld zat verborgen in de kluis.

'The money was hidden in the vault.
b. Teken een lijn in die cirkel.
The place where we would look for the money in (14 a) is the porous
interior of the vault, and not its solid outer shell; the possible location
of the line in (14 b) is the area circumscribed by the geometrical figure
"circle", but not the circular line itself (though this line may be considered
as a limit of the relevant area). In short, in (14) y's boundary does not
enter into a COINCIDENCE relation in with x. It can be observed that
the internal consistency of y's boundary in (14) is different from the
material it circumscribes. As such, the entities y in (14) when seen in
their full-fledged dimensionality can easily be thought of as consisting
of two mutually exclusive parts, viz., the boundary and the interior; and
both parts cannot be simultaneously involved in a COINCIDENCE
relation.
Then, again, the entire entity y (seen in its full-fledged dimensionality)
may be involved in the following instances:
(15)
a. de worm in de appel
'the worm in the apple'
b. de nage I in de plank
'the nail in the board'
c. de melk in de koffie
'the milk in the coffee'
d. de auto in de mist
'the car in the fog'
39
The entities y in (15) can be seen in their entirety because there is little
or no difference in consistency between their boundaries and the material
they circumscribe (15 a-c) or because y does not feature any clear boundaries at all (15d). 16
In sum, in sometimes lexicalizes a COINCIDENCE relation with the
entire entity y (15), sometimes with a portion thereof (14). Following
Hawkins (1988: 251), we will refer to that particular part of y (whether
it be the entire entity y or a portion thereof) that is involved in a
COINCIDENCE relation such as in as the "active zone".
Given that it is not necessarily the entire entity y that enters in the
COINCIDENCE relation expressed by in, it is not the spatial characteristics of y as such that define in (as Clark and Quirk and Greenbaum
wrongly suggest),17 but those of the relevant active zone associated with
y. In what follows, we will refer to such an active zone of y with its
spatial characteristics as a spatial configuration.18 As such, in lexicalizes
a COINCIDENCE relation between and a specific spatial configuration.
What, now, is the spatial configuration characteristic of iri> We will
examine three possibilities: the spatial configuration associated with y is
(a) a 3DIM entity (based on Clark 1973); (b) a 3DIM volume or 2DIM
area (based on Quirk Greenbaum 1973); or (c) an interior (Bennett
1975; Cooper 1968).
(a) Describing in as a COINCIDENCE relation of with a 3DIM
spatial configuration is clearly insufficient. As we have already pointed
out, also 2DIM and 1DIM entities y which, as a matter of course
feature 2/1 DIM active zones can give rise to in.
(b) A spatial configuration characterized as a 3DIM volume or a 2DIM
entity gives a fairly exhaustive picture of in. Still, I have a few reservations
about Quirk and Greenbaum's approach. For one thing, it does not
account for the (admittedly infrequent) use of in with 1DIM entities (cf.
above). More importantly, it does not indicate why 3DIM and 2DIM
entities alike can give rise to in, in other words, what the link is between
3DIM volumes and 2DIM areas. This is all the more important, because
not all 2DIM entities trigger off in:
(16)
Hij zat op de vloer.

'He sat on the floor.'
(c) In Bennett's (1975) and Cooper's (1968) view, inE lexicalizes a

COINCIDENCE relation between and the spatial configuration interior. To my mind, the notion "interior of y" is inextricably linked to the
presence of boundaries in y in that it refers to that particular portion of
40
Hubert Cuyckens
y that is enclosed by its boundaries (e. g., the porous interior of the vault
in (14 a) or the area circumscribed by the circular line in (14 b)).
A semantic description of in along these lines definitely meets the
objection against Quirk and Greenbaum's analysis. Indeed, the notion
"interior" establishes the link between 3DIM volumes and 2DIM areas:
when 2DIM areas, like 3DIM volumes, feature a bounded interior, they
may give rise to in (compare (14 a) and (14 b)). However, equating IN
(x, y) with COINCIDENCE (x, interior(y)) seems too restrictive. Indeed,
I fail to see what sort of interior one would attribute to the entities y in
(17), all of which enter into a COINCIDENCE relation expressed by in.
(17)
a. Wij hebben heel wat vrienden in de buurt.

'We have a lot of friends in the neighborhood.'
b. Ik had die schop in mijn maag niet verwacht.
was not prepared for that kick in the stomach.'
c. Kinderen begraven graag hun vaders in het zand.
'Children like to bury their fathers in the sand.'
Of course, one might retort that these entities y do not have interiors,
but that in some way they are interiors, but this makes the notion
"interior" devoid of all meaning. 19 Indeed, I fail to see how the entity
named by the stomach in (17 b), for instance, can be seen as an interior.
One reason why the incorporation of "interior" in the semantics of in
runs into problems might be that it too strongly suggests the presence of
a boundary in y (cf. Hawkins 1985: 96). It can be observed, though, that
in is non-committal with respect to the existence of such a boundary: the
entities y in (14) have definite boundaries, but in (17) no such boundary
is apparent.
So far, we have analyzed in as a COINCIDENCE relation between an
argument and a specific spatial configuration associated with y. The
descriptions of the spatial configuration we looked at so far all proved
to be inadequate in that none covered all the spatial relations lexicalized
by in. Relatively speaking, Quirk and Greenbaum provide the most
thorough analysis; yet, they do not indicate what links the differing
spatial configurations (3DIM volume, 2DIM area) that all give rise to
in.
3.1.1.2.
Instead of describing in as the combination of the rather abstract relation
COINCIDENCE and the spatial configuration associated with y, one
might also turn to a more specific characterization of the relational
content of in. Here are some definitions of inE and dans:
41
Leech: "/ expresses the concept of'enclosure' or 'containment' as applied

either to two-dimensional or three-dimensional locations" (1969: 162).
Miller and Johnson-Laird: "IN (x,y): A referent is 'in' a relatum y if
... [PART (x, z) & INCL (z, y)]" (1976: 385). Moreover, they state that
the relatum y must be the kind of thing that has an interior.
Vandeloise: "a est dans ... b si le site [i.e., y] contient (partiellement) ...
la cible [i.e., x]" (1986: 224). Somewhat further (p. 225) he specifies this
containment relation as follows: "Le contenu est inclus, au moins partiellement, dans le contenant ou dans la fermeture convexe de sa partie
contenante."
I do not think that a semantic description of in based on the definitions
above (i. e., making use of any one of the relations CONTAINMENT,
ENCLOSURE, or INCLUSION) is drastically different from the previous definitions combining the relation COINCIDENCE with a specific
spatial configuration associated with y. For one thing, CONTAINMENT,
ENCLOSURE, and INCLUSION each entail COINCIDENCE. Furthermore, the extra information (extra, that is, with respect to the abstract
relation COINCIDENCE) that in the previous definitions was relegated
to the spatial configuration of y is here part of the relational information
itself. Consider the following example. A spatial relation between de
juwelen 'the jewels' and de doos 'the box', for instance, can on the
basis of the earlier definitions be lexicalized by means of in because
the box (y) features a 3DIM active active zone, viz., its 3DIM interior.
The semantic representation CONTAINMENT (jewels, box) of IN (jewels, box) signals that the jewels coincide with an active zone of y that
can be described as the containing part of y (possibly the entire entity
y). In sum, then, in both types of definitions, the spatial configuration
present in y is of crucial importance.
Now, I must admit that the characterization of this spatial configuration in the earlier definitions differs from that of the present ones: the
former often included dimensional properties, whereas the latter is expressed in terms of "containing part", "including part", "enclosed part".
Let us now look at each of these definitions.
(a) The semantic representation ENCLOSURE (x, y) or COINCIDENCE between and the spatial configuration "enclosed part of y"
definitely does not cover all the spatial relations lexicalized by in. It
singles out those relations between and a clearly bounded entity y,
which is precisely the type of entity that features a spatial configuration
"enclosed part". In fact, the spatial configuration "interior" experiences
42
Hubert Cuyckens
the same difficulties. As such, ENCLOSURE (x, y) cannot account for

the following instances:
(18)
a. de auto in de mist
'the car in the fog'
b. de oase in de woestijn
'the oasis in the desert'
c. Wij hebben heel wat vrienden in de buurt.
'We have a lot of friends in the neighborhood.'
d. Ik had die schop in mijn maag niet verwacht.
was not prepared to that kick in the stomach.'
e. Kinderen begraven graag hun vaders in het zand.
'Children like to bury their fathers in the sand.'
(b) INCLUSION (x, y), or COINCIDENCE between and the spatial

configuration "including part of y", covers both bounded and unbounded
3DIM entities y, but it seems a less apt description of spatial relations in
between and a 2DIM entity y.
(c) In (19), both the glass and the tree feature a spatial configuration
"containing part", which enters into a COINCIDENCE relation with the
fly and the birds, respectively.
(19)
a. de lieg in het glas

'the fly in the glass'
b. de vogels in de bomen
'the birds in the trees'
This is what led Vandeloise (1986) to put forward his semantic description
of dans, founded on the concept CONTAINMENT, instead of the ones
above. However, CONTAINMENT, like ENCLOSURE and INCLUSION, hinges too much on the presence of boundaries in y, which makes
it a less felicitous characterization of the spatial relations in in (18). In
view of this, it is quite significant that Vandeloise's discussion of dans
largely confines its examples to relations between and bounded entities
y. Indeed, it is only those that are preferably associated with the spatial
configuration "containing part".
In the foregoing, we have reinterpreted CONTAINMENT, INCLUSION, and ENCLOSURE as a COINCIDENCE relation between and
a specific spatial configuration associated with y. Basically, we did this
because an SpP such as in first and foremost serves to locate an entity
43
with respect to y. I think this locative function is best captured by the

abstract relation COINCIDENCE (which is then further supplemented
with extra information about y).
If we do not reinterpret INCLUSION/CONTAINMENT/ENCLOSURE, we encounter an extra problem. At the beginning of this section,
I pointed out that in shares the relational information COINCIDENCE
with other SpPs such as op. Defining in in terms of a specific relation
such as INCLUSION, 20 for instance, compels us to describe op in terms
of another specific relation, viz., SUPPORT. As such, we seem to lose
out on the obvious fact that in and op share the relational information
"COINCIDENCE".
Now, it is not only relational information that in shares with other
prepositions. The spatial configuration of y that features in in also occurs
in a number of other prepositions such as uit 'out of and door 'through':
(20)
a. De vogel is in zijn kooi.

'The bird is in its cage'
b. De vogel is uit zijn kooi.
'The bird is out of its cage.'
c. De trein rijdt door de tunnel.
'The train passes through the tunnel.'
It can be observed that, although the relational information in these

examples varies (COINCIDENCE in (20 a), NON-COINCIDENCE in
(20b-c)), the same 3DIM spatial configuration 21 of y is involved. This
semantic similarity between in, uit, and door can only be revealed when
the spatial configuration of y is an integral part of the definition of in
(in other words, when in is not described in terms of INCLUSION/
CONTAINMENT/ENCLOSURE).
3.1.1.3.
Two conclusions can be drawn from the above discussion. First, preference should be given to a treatment of in in terms of a rather abstract
COINCIDENCE relation between and a specific spatial configuration
associated with y over a description in terms of more specific relational
information.
Second, we have seen that existing descriptions of this spatial configuration (viz., 3DIM, 3DIM volume/2DIM area, interior, enclosed part,
containing part, including part) are not satisfactory. There is no spatial
configuration with one single unitary definition that enters in all CO-
44
Hubert Cuyckens
I N C I D E N C E relations in, but each one is involved in a number of them.

This means that the semantics of in, which crucially hinges on this spatial
configuration, cannot be characterized in necessary and sufficient terms. 2 2
Consequently, I think we should give up on the idea of trying to find
one single unitary (i.e., necessary and sufficient) definition of in. If we
wanted to present " C O I N C I D E N C E between and the spatial configuration interior", for instance, as a necessary and sufficient description
of in, we would have to stretch the term "interior" in order to account
for instances such as those in (18). I admit that the spatial relations in in
(18) are not wholly unrelated to cases such as (11 a) and (14 a), but the
former do not express containment in the same way the latter do. W h a t
we are dealing with, then, is differing yet related uses of in.
In order to capture this, we submit that the spatial configuration of y
that gives rise to in consists of a number of interrelated subconfigurations
which can be described in terms of a n u m b e r of interrelated subsets of
spatial attributes (each subset contains a number of attributes (possibly
one) that is shared by other subsets). M o r e specifically, the spatial configuration characteristic of in can be assigned a family resemblance
structure.
This spatial configuration will be labelled "medium". This term is
introduced by Hawkins (1985: 95), who defines "medium" as "a condition,
atmosphere in which something may function or flourish" (Webster). 23
One might object that this definition is too vague or too general, that it
is in need of further explanation. But that's precisely the point: in
lexicalizes a number of differing, b u t interrelated C O I N C I D E N C E relations, and that is what the term " m e d i u m " is intended to capture.
In sum, we submit that in lexicalizes a C O I N C I D E N C E relation
between and the spatial configuration " m e d i u m " associated with the
entity y.
I N (x, y) = C O I N C I D E N C E (x, medium (y))
The question now is: What spatial (sub-)configurations can be called
"medium"? W h a t are its spatial characteristics, and concomitantly, what
entities y show the spatial configuration "medium"? In this section, we
will restrict ourselves to the different types of medium. 2 4 It is only in
Section 3.3. that we will explore how the different types of medium are
related, that is what makes them into a genuine family resemblance
structure.
3.1.2.
45
What are the different mediums?
3.1.2.1.
Consider the following examples:
(21)
a. de boeken in de kast
'the books in the cupboard'
b. de Juwelen in de doos
'the jewels in the box'
c. Mijn voet zit nu in mijn schoen.
'My foot is now in my shoe.'
d. De stoel Staat in de hoek van de kamer.
'The chair is in the corner of the room.
e. De vogels bouwden een nest in het gat van de muur.
'The birds built a nest in the hole in the wall.'
In (21 a-c), the active zone of y that enters into the C O I N C I D E N C E

relation expressed by in is the interior of y; in (21 d), it is the space
defined by two planes; and in (21 e) it is the hole denoted by y. In each
of these examples, the active zone of y is a 3DIM, porous, bounded space.
It is this active zone that constitutes the first type of medium. Incidentally,
in (21 a-d), the active zone takes up part of y, while in (21 e) it denotes
the entire entity y. Let us now look at this in somewhat more detail.
(a) The first type of 3 D I M , porous, bounded medium associated with
a particular entity corresponds with the interior defined by its solid outer
shell. All entities that can be characterized as "shells or frames with a
solid external boundary circumscribing a three-dimensional interior with
a consistency quite different from that of the b o u n d a r y " (Hawkins 1988:
253) feature this medium. Examples are: cupboard, closet, room, grave,
building, various sorts of smaller containers (bottle, barrel, box), etc.
Now, as Herskovits points out, shell-like entities with complete closure
are relatively rare, "so that some boundaries of the interior will usually
be imaginary" (1986: 151). Obvious examples are:
(22)
a. We zijn nu in de tunnel.
'We are now in the tunnel.'
b. De kerstboom Staat in de woonkamer.
'The Christmas tree is in the living room.'
Two sides of the interior of a tunnel (22 a) are planes through its ends.
Rooms in a house (22 b) may be separated from each other by an
imaginary line instead of solid walls. Also the entities y (with incomplete
46
Hubert Cuyckens
Figure 5. D e vis in zijn hand
closure) in the following instances may be attributed an interior or, more

generally, a 3DIM, porous, bounded medium:
(23)
a. de vis/de pen in zijn hand

'the fish/the pen in his hand'
b. de voet in de stijgbeugel
'the foot in the stirrup'
c. de vogel in de boom
'the bird in the tree'
d. de aardappels in de kom
'the potatoes in the bowl'
The medium that bears a COINCIDENCE relation with the fish or the
pen in (23 a) is a porous, 3DIM space bounded by the palm of the hand
and a number of imaginary planes through its edges (see Figures 5 and 6).
Similarly, the medium defined by a torus-shaped object such as stirrup
(23 b) is an interior bounded by the material object itself and imaginary
planes through its sides. In (23 c), the interior (and medium) of the tree
Figure 6. D e pen in zijn hand
47
is the 3DIM space bounded by its outline. Entities with partial closure
also typically include open containers and vessels such as vase, glass, cup,
etc. (cf. also 23 d). Their medium consists in the 3DIM, porous active
zone that is bounded by a closed bottom, a vertical wall, and an imaginary
plane through the rim of the vessel or open container. Along similar lines,
a medium may be discerned in "cup-like concavities" (cf. Herskovits
1986: 149) such as cradle, pocket, nostril, armchair, etc.
It is important to keep in mind that not all 3DIM spaces bounded by
(the external boundaries of) an entity with incomplete closure qualify as
an interior (and hence as a medium), or qualify as such in every context.
Compare the following situations:
(24)
a. de aardappelen in de kom
b. de aardappel onder/*in de kom
'the potato under/*in the bowl'
Both in (24 a) and (24 b) the entity y bowl defines a 3DIM porous space
bounded by its bottom and vertical walls and an imaginary plane through
its rim. Yet, only in (24 a) does this space constitute the active zone of a
COINCIDENCE relation in (with x); in other words, only in (24 a) does
it qualify as an interior (and correspondingly, as a medium).
Why do objectively identical spaces qualify as a medium in one case
and not in the other? Ultimately, the answer to this question rests on
whether the 3DIM, porous, bounded space at issue is considered part of
the entity it is circumscribed by or not. In that regard, it might be
instructive to look at shell-like entities with complete closure again. These
entities function as containers in that they can hold, carry, or store things.
It seems self-evident that their containing part (viz., the 3DIM, porous
space or interior circumscribed by the container's outer shell) is an integral
part of these entities. As a part of y, then, their interior constitutes (in a
COINCIDENCE relation with x) the type of active zone of y that we
have labeled "medium". By analogy, those shell-like entities with partial
closure that function as containers (cf. 24 a) also circumscribe the type
of 3DIM, porous, bounded active zone that we labeled "medium".
In (24 b) the bowl is turned upside down, which precludes its functioning as a proper container (conceptually it is no longer fit to hold, carry,
or store things, though materially of course nothing would prevent this).
In cases such as these, no containing part can be attributed to y; or, to
put it differently, the 3DIM porous space defined by the bowl (turned
upside down) is not considered a part of y. As such, it cannot serve as
48
Hubert Cuyckens
the type of active zone that meets the characteristics of a medium. Other
examples illustrating the importance of y's functioning as a container:
(25)
(26)
de rook in de kaasstolp
'the smoke in the cheese cover'
a. de melk in het glas
'the milk in the glass'
b. de melk op/*in het glas
'the milk on/*in the glass'
(27)
Afspraak onder/*in de Arc de Triomphe.

'Meet us under/*in the Arc de Triomphe.'
(28)
Wij zaten met onze benen onder/*in de tafel.

'We had our legs under/*in the table.'
All entities with incomplete closure in examples (24)-(28) involve vessels,

open containers, or cup-like concavities, though of course only cheesecovers and glasses really function as containers. Now compare the following situations featuring a torus-shaped object:
(29)
a. De voet in de stijgbeugel (Figure 7)

'the foot in the stirrup'
b. Het muntstuk op/*in de stijgbeugel (Figure 8)
'the coin on/*in the stirrup'
Both in (29 a) and (29 b), the stirrup circumscribes a 3DIM, porous space,
but only in (29 a) is this space seen as its containing part, because only
in (29 a) is de stijgbeugel actualized as a container. Indeed, containers not
only hold, carry, or store things (cf. above), but in doing so, they also
control the position of the contained. Stirrups exert physical control over
49
one's foot in that they hold one's foot in place. As there is no such
relation of physical control between the stirrup and the coin in (29 b),
the stirrup cannot properly be called a container; hence it has no containing part, which in turn implies that the 3DIM, porous space the
stirrup defines cannot constitute an active zone, and medium, of y.
Summing up, the first type of 3DIM, porous, bounded medium that
we looked at in (a) is the interior of shell-like entities with complete or
partial closure. It was shown that the presence of an interior in such an
entity is inextricably linked to its functioning as a container; 25 and this
functioning, in turn, is linked with there being some form of control from
the physical part of the container. 26
(b) A second type of 3DIM, porous medium involves spaces bounded
by two planes or two cylinder-like objects meeting at an angle. Here are
some examples:
(30)
a. De man lag te slapen in de goot.

'The man was sleeping in the gutter.'
b. de stoel in de hoek van de kamer
'the chair in the corner of the room'
In (30 b), for instance, the active zone (and medium) of y in the COINCIDENCE relation in with is the 3DIM, porous space bounded by
two walls of the room. As in (a), these boundaries are not included in
the active zone; that is, the medium attributed to y only represents part
of y.27
(c) The third type of medium denotes the entire entity itself. In these
examples, the active zone of y is a 3DIM, porous space, bounded by
50
Hubert Cuyckens
another object: the frame of the door in (31 a), the wall in (31 b), the
cheese in (31c).
(31)
a. Hij stond in de deuropening.

'He was standing in the door.'
b. de muis in het gat in de muur
'the mouse in the hole in the wall'
c. Er zitten veel gaten in dit soort kaas.
'There are lots of holes in this type of cheese.'
In this first section, we have looked at three (slightly differing) types of

3DIM, porous, bounded mediums. Not all mediums need to possess these
three characteristics.
3.1.2.2.
(a) Consider the following examples:
(32)
a. de nagel in de balk
b. Hij werd levend begraven in de muur.
'He was buried alive in the wall.'
c. het gatjde barst in de muur
'the hole/the crack in the wall'
In these examples, the active zone of y that enters into a COINCIDENCE

relation with is the entire solid chunk of material denoted by y, or more
generally, the 3DIM, non-porous, bounded portion of y. It is this active
zone that makes up the second type of medium. The entities that feature
this particular medium are solid chunks of material such as plank 'board';
muur 'wall'; paal 'pole', etc. Examples such as (32 c) should not pose any
special problems: just like any other material object, a hole, gap, or crack
may bear a COINCIDENCE relation with a 3DIM, solid medium associated with y.
Unlike in the first type of medium (a porous, 3DIM space bounded
by solid material), there is no contrast in the internal consistency of this
medium and its boundary: both consist of solid matter. Note, furthermore, that, although the active zone or medium associated with y in (32)
takes up the entire entity y (i.e., inclusive of its free outward facing
boundaries), these boundaries are not highlighted. Note, however, that
when they are free, outward facing surfaces, y's boundaries may be
actualized as a separate active zone, which enters in a COINCIDENCE
relation expressed by op. Compare:
(33)
a. De nagel
'The nail
b. De nagel
'The nail
51
zit in de plank.
is in the board.'
ligt op de plank.
is on the board.'
We stated above that the entities featuring this type of medium are solid
chunks of material. This solid chunk may exist on its own (cf. 32), or it
may be a shell circumscribing an interior. As such, the following expression is ambiguous:
(34)
de nagel in de kast
'the nail in the cupboard'
Two active zones can be discerned in the entity y, viz., its interior (3DIM,
porous, bounded) and its outer shell (3DIM, non-porous, bounded), both
of which qualify as mediums and, as a result, may give rise to in.
(b) Not every 3DIM, non-porous active zone of an entity corresponds
to a solid chunk of material. Among the entities featuring a 3DIM,
bounded, non-porous medium, also less solid substances occur, which
need to be kept in a container so as not to spread incontrollably.
(35)
a. Er zit een muis in de haver.

'There is a mouse in the oats.'
b. Ik heb nog geen suiker in mijn koffie.
didn't put any sugar in my coffee yet.'
c. de stenen in de plas
'the stones in the puddle'
d. de kruimels in de jam
'the crumbs in the jam'
The less solid entities exemplified in (35) can hardly feature a separate
active zone giving rise to op. Precisely because they are less solid and
therefore need to be contained, they do not show a free outward surface.
And even if they did (e. g., the top surface in (35 b)), this surface may
not be solid enough to support another entity.
3.1.2.3.
Look at the following examples:
(36)
a. Zij zaten met hun benen in het zand.

'They were sitting with their legs in the sand.'
b. Alle vissen zwemmen in het water.
'Fish swim in water.'
52
Hubert Cuyckens
Although objectively the entities in y denote an entire body of sand (on

the beach, for instance) or water (the sea), conceptually y only equates
with a relevant portion thereof without clear boundaries. The active zone
of y in the C O I N C I D E N C E relations above corresponds with one such
chunk of substance. This third type of medium, then, can be characterized
as 3DIM, non-porous, and unbounded. Given that the entire entity y has
conceptually been reduced to one relevant chunk, also this medium
corresponds to the entire y.
The relevant chunk of substance denoted by y may also combine clear
and vague boundaries. In (37), the relevant chunk of water has two clear
boundaries, viz., its outward top surface and its bottom surface.
(37)
de vissen in het water

'the fish in the water'
3.1.2.4.
In (38) the active zone of y, and concomitantly its medium, denotes the
entity y itself, which is a 3DIM, porous, unbounded space. Allowance
should be made for the fact that, if y designates a large body of space,
it may reduce to a relevant chunk with vague boundaries.
(38)
a. het vliegtuig in de mist j de wlken

'the plane in the fog/the clouds'
b. de vogels in de lucht
'the birds in the air'
c. Er zijn zwarte gaten in het heelal.
'There are black holes in the universe.'
d. Grootvader is nu in de hemel.
'Granddad is now in heaven.'
3.1.2.5.
So far, we have dealt with 3DIM mediums with variable specifications
for the criteria "boundedness" and "internal consistency". Mediums may
also vary along the dimensionality-criterion, that is they may also be
2DIM. Consider the following situations:
(39)
a. De kinderen stonden in de cirkellhet vierkant.1%

'The children were standing in the circle/the square.'
b. Teken een cirkeltje in de hoek gevormd door de twee rechten.
'Draw a circle in the corner formed by the two lines.'
53
c. Je kwotering Staat in de marge.

'Your marks are in the margin.'
d. Hij woont in Duitsland.
'He lives in Germany.'
In all these examples, the active zone, and medium, of y is a clearly
bounded, 2DIM area. Now, what entities exhibit this type of medium?
(a) A first set, exemplified in (39 a), includes geometrical entities (circle,
square, rectangle, etc.). These 2DIM areas are bounded by a 1DIM line,
which itself is not part of the medium. Indeed, this 1DIM bordering line
constitutes an active zone of its own (De kinderen stonden op de cirkel
'The children were standing on the circle'). (Notice the similarity with
3DIM entities with complete closure.)
(b) The entity y in (39b) denotes a set of lines meeting at an angle and
circumscribing a 2DIM area. As in (a), these boundary lines are not part
of this type of 2DIM-area medium, but they do not make up an active
zone by themselves either (see also note 27).
(c) The entity y in (39c) designates an intrinsic part of a page, bounded
by a(n) (imaginary) line marking the division of the page. Again, this
2DIM medium does not include the dividing line (if it is materially
present).
(d) A large set of entities featuring a 2DIM, bounded area is geographical and geopolitical entities (cf. 39 d). Conceptually, we think of cities,
countries, and the like as 2DIM areas bounded off from the rest of the
earth. Just like entities such as circle, the boundaries of geographical/
political entities can constitute an active zone of their own ( H i j woont op
de grens 'He lives on the border').
3.1.2.6.
In the previous section, we have seen that 2DIM, bounded mediums may
be attributed to 2DIM entities with boundaries that clearly set them off
from the rest of the surrounding area or environment. There are also
2DIM entities (geographical or geopolitical entities) which are marked
off from the rest of the earth's surface, but not in a well-delineated way.
Also these vaguely bounded, 2DIM areas may constitute a medium:
(40)
a. Wij wonen in het noorden/de woestijn/de prairie.

'We live in the north/the desert/on the prairie.'
b. Er is hier geen Supermarkt in de buurt.
'There is no supermarket in the neighborhood.'
54
Hubert Cuyckens
c. Wij hebben een buitenverblijt in de bergen.

'We have a summer house in the mountains.'
3.1.2.7.
Let us, by way of conclusion, sum up the different types of medium and
provide a few extra examples.
(a) 3DIM, bounded, porous medium:
(41)
a. Er ligt veel stof in de kast.

'There is a lot of dust in the cupboard.'
b. Laten we afspreken in de luchthaven.
'Let us meet in the airport.'
c. De kinderen zitten in de auto.
'The children are in the car.'
d. We zijn in de
kamer/slaapkamer/badkamer/toilet/kelder/keuken/
hall.
'We are in the room/bedroom/bathroom/basement/kitchen/
hallway.'
e. Hij woont in een villaatje.
'He lives in a country house.'
f. Hij woont in onze straat.
'He lives in our street.'
g. Zij heeft een kind in haar armen.
'She's got a child in her arms.'
(b) 3DIM, bounded, non-porous medium:

(42)
a. Hij voelde een scheut van pijn in zijn vinger.

'He felt a twinge in his finger.'
b. Hij heeft een ring in zijn oor.
'He has got a ring in his ear.'
c. De ketel hangt in het vuur.
'The kettle hangs in the fire.'
d. Het goud zit in het erts.
'The gold is in the ore.'
(c) 3DIM, unbounded, porous medium (cf. (38))

(d) 3DIM, unbounded, non-porous medium:
(43)
a. Jezus wandelt in het water.

'Jezus walks in the water.'
55
b. We zitten in het stof.

'There is dust all over the place.'
(e) 2DIM, bounded medium:
(44)
a. Zij heeft sproeten in haar gezicht.

'She has freckles in her face.'
b. De verhalen doen de ronde in het dorp.
'The story is being told all over town.'
c. Zij wonen in de stad.
'They live in town.'
d. London ligt in Engeland.
'London is in England.'
e. Het elftal Staat in het veld.
'The team is out in the field.'
f. Hij werkt in de haven.
'He works in the harbour.'
(f) 2DIM, unbounded medium:

(45)
Hij is alleen in deze wereld.

'He is all alone in this world.'
So far, we have abstracted from the uses of in that indicate a part-whole

relationship. In the following section, we will now explore this somewhat
further.
3.2. "In" in part-whole relationships

Let us first consider a few examples of part-whole relationships lexicalized
by in:
(46)
a. de spieren/de botten in zijn been

'the muscles/the bones in his leg'
b. de golving in de weg
'the undulation in the road'
c. een bocht in de weg
'a curve in the road'
d. de knik in de curve
'the dip in the graph'
If we consider cases such as (46 a) only, one might be misled to think

that part-whole relations lexicalized by in, analogous to spatial relations
56
Hubert Cuyckens
between two discrete objects, denote a COINCIDENCE relation between

the part ( = x) and the medium associated with the whole ( = y). In (46 a),
this medium would be the 3DIM, non-porous, bounded active zone that
is denoted by y. Now, the spatial configuration "medium" can hardly be
assigned to the entities y in (46 b-d): weg in (46 b) is a 2DIM surface; in
(46 c) weg is a 1DIM line; also in (46 d), curve is a 1DIM line. As such,
in contrast to in expressing a NPW-relation, in in PW-relations does not
lexicalize a COINCIDENCE relationship with the medium associated
with y. In the next section, we will see, though, that there is a straightforward explanation for PW-in.
3.3. The family resemblance structure of "in"

We will first describe the family resemblance structure of NPW-/, which
crucially involves the concept "medium". We will then explain how PWin hooks onto this structure.
3.3.1.
We have seen that the concept "medium" cannot be described in terms
of a set of necessary and sufficient conditions. On the classical view, the
failure to provide a single set of criterial features for a lexical concept
would automatically imply that the various featural configurations associated with that concept should be considered as homonymous, and
hence unrelated uses of the concept. Now, the various featural characterizations of "medium" that we have given are not unrelated, but they are
connected to each other along various dimensions, very much like the
different members of a family (some featural configurations share the
feature "3DIM", others the feature "bounded", etc.). In other words, the
various instances of the concept "medium" are related to each other, not
on the basis of their identity, but of their mutual similarity. In short, the
concept "medium" can be ascribed a family resemblance structure.
Let us first look at the question how the different uses of "medium"
are related, in other words, at what the synchronic connections in the
language user's conceptual knowledge about "medium" are.
The first type of medium we distinguished was characterized as the
3DIM, porous, bounded active zone of y, with a containment function
(e.g., de sleutels in de hast 'the keys in the closet'). This medium relates
(a) to 3DIM, non-porous, bounded mediums (e. g., de nagel in de plank
57
'the nail in the board') and (b) to 3DIM, porous, unbounded mediums
(e. g., We liepen in de mist 'We were walking in the fog). With the former,
it shares the features "3DIM" and "bounded", with the latter, it shares
the features "3DIM" and "porous". The fact that there exist 3DIM
mediums which are either non-porous or non-bounded motivates the
existence of 3DIM mediums which are non-porous and non-bounded at
the same time (e. g., met zijn hoofd in het zand 'with his head in the sand').
The feature "boundedness", which so far has gone together with threedimensionality, lies at the basis of another conceptual extension. Indeed,
bounded 2DIM mediums may also give rise to in (e. g., in de cirkel 'in
the circle', in de marge 'in the margin', in Engeland 'in England'). These
2DIM bounded mediums must mark some division. As Herskovits correctly points out, "the reference object must be one of several areas
arising from dividing a surface" (1986: 153). These can be divisions on a
page, geometrical figures, geographical and geopolitical entities, and
divisions on the human body.
The third and the fourth types of division lie at the basis of one more
conceptual extension: bounded divisions on the earth's surface motivate
unbounded divisions on the earth's surface (e. g., Hij woont in Ohio 'He
lives in Ohio' Hij woont in de vallei van de Ohio rivier 'He lives in
the valley of the Ohio'). Similarly, bounded divisions of the human body
can motivate unbounded divisions (e.g., Ze sloegen hem in zijn gezicht
'The hit him in the face' > Hij kreeg een schop in zijn maag 'They kicked
him in the stomach'). We see here that it is only a subset of entities
featuring a particular medium that motivates another medium (divisions
on the earth's surface motivate unbounded divisions), whereas it is, for
instance, the entire category of 3DIM bounded mediums which motivates
bounded 2DIM mediums.
At this point, we have arrived at the edge of the category or concept
in. Not every entity featuring a bounded or unbounded medium that
marks a division on the earth's surface gives rise to in. Names of islands,
for instance, often take op. Op is the logical alternative here because it
lexicalizes, among other things, the relation between and a 2DIM
surface. The choice between in and op may not be as erratic as it seems,
though. It seems to me that the more an island can be thought of as a
political unit (just like a country on the continent, with well-delineated
boundaries), the more the boundedness of the island comes into play and
the higher the likelihood of in. Furthermore, unbounded divisions of the
human body are preferably lexicalized by means of op. It is indicative
that at the edges of categories, languages differ from each other. For
58
Hubert Cuyckens
instance, English allows in for unbounded divisions of the human body,

whereas in Dutch we largely have op. Another conclusion we can draw
from the variation at category edges is that it is possible to explain the
particular use of an SpP (e. g., in in een schop in de maag 'the kick in the
stomach'), but it is not always possible to explain why the same SpP is
not used in similar cases (e. g., Hij kreeg een schop op zijn borst vs. He
was hit in the chest).29
By way of summary, Figure 9 represents the family resemblance structure of the spatial configuration "medium", in which the relations between
the different types of medium have been rendered through shared features.
In a family resemblance structure (a network of overlapping featural
configurations), each of the configurations need not share a number of
features with all of the other configurations. A family-resemblance structure may also consist of a chain of related configurations. In the latter
setup, a particular configuration at one end of the chain may not share
any features with another configuration at the other end of the chain;
that is, these two configurations are only indirectly related through the
other intermediary configurations.
1 = "3DIM"
2 = "bounded"
3 = "porous"
4 = "non-porous"
5 = "vaguely bounded"
6 = "2DIM"
Figure 9. The family resemblance structure of "medium"
59
We see that the concept "medium" is a mixture of these two types of

family resemblance structure. Each of the configurations which share the
feature "3DIM" is related to all of the other 3DIM-configurations.
Chaining occurs in the extension from bounded, 3DIM to bounded,
2DIM mediums. Both configurations share the feature "boundedness",
but the bounded, 2DIM mediums are not (directly) related to the other,
unbounded, 3DIM mediums. 2DIM, unbounded mediums are chained
to 2DIM bounded ones, with which they share the feature "2DIM division
of the earth's surface". They are only indirectly related to 3DIM mediums
through the intermediary link of 2DIM bounded divisions of the earth's
surface.
3.3.2.
How can we explain the use of PW-in on the basis of the family resemblance structure for NPW-in? Let us consider the following examples of
non-part-whole in again:
(47)
a. de spijker in de plank
'the nail in the board.'
b. Ik wil graag melk in mijn koffie.
would like cream in my coffee.'
In our earlier discussion of in (Section 3.1.), we pointed out that in in

(47) expresses a COINCIDENCE relation between and a 3DIM,
bounded, non-porous medium associated with y. What we did not point
out then is that such a COINCIDENCE relation conveys that is
embedded in y. I think it is this "embeddedness" between and y that
motivates the use of in in the part-whole relations in (46). The undulating
part of the road is embedded in the rest of the road (46 b); the curving
part of the road is embedded in the rest of the road (46 c). Note that not
all part-whole relationships are lexicalized by means of in:
(48)
a. de leuning aan de stoel

'the back of the chair'
b . de Striemen
op zijn
buik
'the streaks on his belly'

Figure 10 shows how the part-whole uses of in are related to its nonpart-whole uses.
60
Hubert Cuyckens
3.4. Fuzziness
We have established so far that in can be represented as a family resemblance structure. Now, the fact that in is of a family resemblance nature
is not all there is to say about the semantics of in. Indeed, in is also a
fuzzy concept.
To start with, some subsets of concept instances of "medium" have
fuzzy boundaries. For instance, there is the fuzziness of 3DIM containers
that are open at the top. Containers such as glass unmistakably ask for
in. But, the more the vertical sides of a container decrease in size, the
more one might be inclined to use op instead of in. It seems to me that
as long as one sees y in its containing function, in is allowed. Consider
the following instances. When one sits in the back of a pick-up truck,
one can still sit in the pick-up. Op, however, is also allowed when one
highlights the flat bed ( = 2DIM surface) of the pick-up. Entities with
negligible vertical sides can hardly be assigned a containment function;
therefore COINCIDENCE relations with entities y such as ondiep bord
'dinnerplate' and schotel 'tray' are most appropriately lexicalized by
61
means of op. Still, because of the slightly raised sides of dinnerplates and
trays, I feel I cannot exclude in in (49 a):
(49)
a. Er lag een aardappel op I in het bord.

'There was a potato on/in the plate.'
b. Eet de soep inj*? op je bord eerst op.
'Eat the soup in your plate first.'
Although the vertical sides of a soup plate (49 b) are only slightly higher
than those of a dinner plate, soup plates are typically seen as containers
for soup and other liquids.
Second, fuzziness is also inherent in the semantic element "COINCIDENCE". Indeed, this COINCIDENCE may not be complete. It is
complete in cases such as (50 a), but it is not in (50 b-d).
(50)
a. Mijn pen ligt in de lade

'My pen is in the drawer.'
b. Hij heeft een ring in zijn oor.
'He has got a ring in his ear.'
c. Hij is in het water.
'He is in the water.'
d. Je loopt voortdurend in het stof.
'You keep running through the dust.'
As can be observed from (50 b-d), the COINCIDENCE can be very

partial. Whether we allow partial COINCIDENCE to be lexicalized by
in or not may be context-determined. Sentence (51)
(51)
De appel is in de kom.
'The apple is in the bowl.'
perfectly describes a situation where is on top of other fruit, but

technically outside the bowl (Figure 11). It is only in contexts where one
Figure 11. De appel is in de kom
62
Hubert Cuyckens
wants to be very precise about the location of that particular piece of

fruit that in is not an appropriate lexicalization. When in (50 c), is
standing with his/her feet in the water because is scared of swimming,
then good swimmers would not say that is in the water.
3.5. The prototype of "in"

Prototype effects in a lexical concept may result from different sources.
(a) One major source of prototypes is the degree of family resemblance
in a concept: it is those concept instances that show the greatest featural
overlap with other instances, and least with those of neighboring concepts,
that can be considered the prototypical concept instances. Now, it seems
rather straightforward to isolate protoypical information in a concept,
or a prototypical concept instance, as long as we are dealing with (i)
concepts whose constituent featural configurations are related to all of
the other featural configurations (concepts of the type bird would be an
example, cf. Geeraerts 1989: 67), or (ii) with chained concepts whose
structure is still relatively simple (e.g., vers (Geeraerts 1987: 283) and
climb (Fillmore 1982)). In those cases with more complex chaining structures (e.g., over, op), isolating a prototype on this basis of maximal
featural overlap might not be straightforward (cf. Cuyckens 1991).
In the concept "medium", chaining is fairly limited (it only occurs
between 3DIM and 2DIM instances), so it might still be possible to
isolate a prototype based on maximal featural overlap. The 3DIM,
bounded mediums are probably the most prototypical. Within the set of
3DIM concept instances alone, 3DIM bounded mediums cannot be
isolated as the most prototypical (their featural overlap is the same as
the 3DIM, non-bounded variant). But, given that 3DIM, bounded mediums also overlap with 2DIM, bounded mediums and that the link
between 3DIM, unbounded and 2DIM, unbounded mediums is only
indirect, 3DIM, bounded mediums are those with the highest family
resemblance, and hence are a good candidate for prototypicality.
Second, prototypical instances in a family resembance structure are
also those that share the least features with other concepts. This makes
all entities featuring a 2DIM medium less prototypical because 2DIM
entities may also give rise to op.
(b) On the basis of the degree of family resemblance alone, porous and
non-porous 3DIM bounded mediums are equally prototypical. Yet intuitively, porous 3DIM bounded mediums seem more prototypical. I tend
63
to agree with Hawkins (1985: 312), who suggests that: "because the
human being ... functions and (sometimes) flourishes in a very porous
medium [i. e., air], and because a large majority of physical objects with
which the human being comes into contact also function in the same
porous MEDIUM, ... MEDIUM configurations with a very porous
internal consistency have a certain primacy over [i. e., are more prototypical than] those with a more solid internal consistency."
(c) In addition to its degree of family resemblance, there are also other
sources of prototypes for the concept "medium". Earlier, we have seen
that there is fuzziness with respect to 3DIM containers that are open at
the top: the boundaries of the concept "medium" grow fuzzier as the
vertical sides of these containers decrease in size. Those instances that
least give rise to hesitancy are more prototypical than others. As such, a
glass is a more prototypical medium than a plate.
(d) A final source of prototypes turns on the relational information in
in. As amply pointed out, in lexicalizes a COINCIDENCE relation
between and the medium associated with y. In this respect, relations
involving complete COINCIDENCE are more prototypical than those
with incomplete COINCIDENCE.
When we combine these four sources of prototypes for the SpP in, it
should be clear that spatial relations with complete COINCIDENCE
between and a 3DIM, bounded, porous medium with relatively high
vertical sides are most prototypical. 30 Examples are:
(52)
a. de melk in hei glas

'the milk in the glass'
b. de juwelen in de doos (Figure 12)
'the jewels in the box'
Figure 12. De juwelen in de doos
64
Hubert Cuyckens
4. "In" as a path-SpP
In the following set of examples, in does not assign to one static or
invariable location, but to a series of locations or a path with respect to
y. More specifically, in denotes a path starting at an unspecified point in
space and whose terminal location is the medium associated with y.
(53)
a. Hij gooide zijn materiaal in de doos.

'He threw his tools in the box.'
b. Hij sloeg een nage I in de plank.
'He drove a nail into the board.'
c. De struisvogel stak zijn kop in het zand.
'The ostrich put its head in the sand.'
d. Het vliegtuig ging de lucht in.
'The plane went up into the sky.'
e. We reden Amsterdam in.
'We drove into Amsterdam.'
f. We reden de woestijn in.
'We drove into the desert.'
Notice that the types of medium that we distinguished in our discussion

of the non-path uses of in also show up here.
How do the path-uses of in complement the family resemblance structure that we have established for non-path uses (both part-whole and
non-part-whole) so far? First, the set of path uses of in as a whole is
transformationally 31 linked to its non-path uses through metonymy. As
Taylor (1989: 127) points out, "there is a natural, metonymic relationship
between the path followed by a moving entity, and one of the infinite
number of points located on the path." Furthermore, non-path-/ has to
do with a particular static relation between and y, whereas path-/
denotes a trajectory (associated with x) at whose terminal point bears
the static relation with y lexicalized by non-path-/. As such, the path
and non-path uses of in are interrelated through a mixture of both shared
and transformationally linked conceptual information. In our representation of the family resemblance of in, an arrow indicates transformationally linked senses, while a regular line signifies a link based on shared
conceptual information.
Figure 13 represents the path- and non-path uses of in. Finally, we
should consider (54):
(54)
Hij woont diep het bos in.

'He lives deep into the woods.'
Figure 13. The path and non-path uses of in
65
66
Hubert Cuyckens
Figure 14.
Although in in (54) assigns to one static or invariable location, is

situated at the terminus of a mental path starting at a vague point outside
the forest and moving into the forest. Therefore, this use of in should be
seen as an extension from the path-uses of in and not from its non-pathuses. (Actually, this sentence is similar to cases such as Hij woont voorbij
het postkantoor 'He lives past the post office.') This use of in is rather
restricted: it seems to require an adverb indicating that the terminus of
the mental path present in in is relatively far removed from y's edges.
Concomitantly, it is probably most acceptable for relations between
and relatively large mediums:
(55)
a. Hij woont heel ver Duitslandjde woestijn in.

'He lives far into Germany/the desert.'
b. Hij woont net het bos in.
?'He lives just into the woods.'
c. * Hij woont Duitsland in.
*'He lives into Germany.'
d. *Mijn juwelen liggen ver de la in.
*'The jewels are far into the drawer.'
Figure 14 is a pictorial or image-schematic representation of how this

use of in relates to the other path-uses of in under (54). (The mental path
is indicated by means of a broken line.)
5. Conclusion
In this cognitive-semantic analysis of the Dutch SpP in, I explored the
following issues. First, the conceptual information in in, along with that
in other SpPs, is broadly encyclopedic in that it is part of, and needs to
be understood against the background of, broader cognitive structures.
Second, I presented a detailed sketch of the family resemblance structure
67
of in, which does far more justice to the semantics of in than the "classical"
set of criterial features. Finally, I argued that some concept instances of
in are more prototypical than others (viz., spatial relations with complete
COINCIDENCE between and a 3DIM, bounded, porous medium with
relatively high vertical sides).
Notes
1. The argument y corresponds to that part of the prepositional constituent the preposition
combines with, while is made up of one or more elements from the rest of the sentence
or the neighboring discourse that contains the head of the constituent the prepositional
constituent is a complement or an adjunct to.
2. It should be kept in mind, though, that these terms also designate the entities involved
in a relation that is expressed by lexical items other than spatial prepositions (cf.
Langacker 1987). Furthermore, they are inextricably linked to the notions "figure" and
"ground" in that in a spatial relation the landmark can be seen as the ground for the
trajector ( = figure) (cf. Hawkins 1985: 48).
3. It would be a misconception to think that only word meaning in cognitive semantics
has a conceptual, and hence mental, status. Also on the classical view, word meaning
is attributed a conceptual status (cf. Saussure 1916:156-157; Trier's term "Begriffsfeld";
Katz 1972: 38). The reason for this ill-conceived, narrow interpretation probably lies
with the fact that it is only in psycholinguistic studies of word meaning and, derivatively,
in the cognitive paradigm in linguistics that the mental (i. e., conceptual) status of word
meaning has been so clearly brought to the fore.
4. Classical lexical semantics encompasses (a) European componentialism (with Pottier
1963; Greimas 1966; Coseriu 1964, 1966; Geckeler 1971); (b) the tradition originating
with Katz and Fodor's seminal paper "The structure of a semantic theory" (1963),
which also includes the meaning postulate approach and a number of psycholinguistic
studies; and, (c) somewhat less importantly, Lyons' relational method (1968, 1977).
Although these studies make up a rather disparate set, they can be subsumed under
one common denominator on the basis of their similar approach to the description of
word meaning.
5. A more detailed treatment can be found in Cuyckens (1991), which, in turn, is largely
based on Geeraerts (1986, 1988).
6. The category of "birds", for instance, is evidently not completely identical, but the
classical approach abstracts from the differences between instances, and only has eyes
for its common features.
7. For completeness' sake, it should be pointed out that cognitive lexical semantics does
not claim that all concepts show a family resemblance structure. Technical concepts,
for instance, ("odd number"; "even number"; "plane geometry figure") are perfectly
amenable to a necessary and sufficient description.
8. The fact that a cupboard, for instance, may function as a container is not just conceptual
information determined by the language system, but reflects our interaction with this
object in the outside world.
68
Hubert Cuyckens
9. Also a structuralist such as Bennett cannot but allow this encyclopedic information in
his semantic description of in front of (cf. Bennett 1975: 83).
10. As a non-path SpP, in assigns (or an element in the semantic structure of x) to a
location with respect to y that is essentially static or invariable.
11. In as a path-SpP denotes a continuous sequence of locations (associated with x) relative
to y. These locations, when taken together, constitute a specific path or trajectory, i. e.,
a spatial configuration with significant extension in one single dimension.
12. For a detailed analysis of op and aan, cf. Cuyckens (1991).
13. Indeed, there is significant semantic overlap between the Dutch in and its cognates.
14. The term "coincidence" is mine. The authors cited make use of the terms "location"
and "place".
15. When, in a COINCIDENCE relation between and y, the spatial/dimensional properties of y are (subjectively regarded as) irrelevant, conceptually autonomous parts
cannot be isolated in y. Evidently, the COINCIDENCE between and y may (objectively) only bear on a particular part of y, but when y is not seen in its full-fledged
dimensionality, this particular part is not subjectively highlighted as a conceptually
autonomous part of y, and hence does not trigger off the SpP which typically lexicalizes
COINCIDENCE relations with that particular portion. For instance, in Meer informatie kan je verkrijgen op het postkantoor 'You can get more information at the postoffice', normally only the interior of the post-office is involved in the COINCIDENCE
relation between and y; yet, when this interior part is not hightlighted, op and not in
is the appropriate lexicalization.
16. It should be kept in mind that in a number of these entities (especially solid ones with
free, outward facing boundaries), the outward facing boundaries constitute, next to the
entity y seen in its entirety, a separate portion of y. (In that respect, these entities
largely resemble the entities y in (14).)
17. If we accept, with Clark and Quirk and Greenbaum, that in depends on the spatial
characteristics of y as such, then the semantic representation COINCIDENCE (x,
3DIM (y)) perfectly applies to the spatial relation between the clothes and the top of
the dresser in De kleren liggen op de kast 'The clothes are on the sideboard', yet this
relation is lexicalized by op and not by in.
18. "Active zone" and "spatial configuration" are related notions: the "active zone" associated with y denotes that portion of y (possibly y itself) that is involved in a spatial
relation with x, whatever its spatial characteristics. A "spatial configuration" associated
with y is an active zone of y with particular spatial characteristics.
19. Unless one assigns a family resemblance structure to "interior" in order to capture its
various uses, but then, again, there are less misleading terms than "interior" (cf. Section
3.1.2.).
20. For the sake of the argument, we assume that INCLUSION provides a satisfactory
account of in.
21. Below, we will refine the characterization of the spatial configuration.
22. Note that Vandeloise (1986) does not present his account of in in terms of the notion
"containment" as a necessary and sufficient one. Yet, the fact remains that his description is insufficient; hence "containment" is not a good candidate for a single unitary
definition of in.
23. Elsewhere, he states that it is something that has the capacity or potential of encompassing or enveloping other things. This definition may be somewhat misleading in
that it focuses on 3DIM configurations.
69
24. "Medium" (in double quotation marks) will refer to the concept or category "medium".
Medium (without quotation marks) will refer to (an) instance(s) of that concept.
25. At this point, due credit should be given to Vandeloise (1986), who first pointed out
that functional considerations play an important role in SpP-semantics. However, as I
pointed out earlier (cf. also note 22), the notion CONTAINMENT alone does not
satisfactorily describe the semantics of in.
26. Indeed, in a sentence such as De kaas ligt onder/*in de stolp 'The cheese is under/*in
the cheese cover', onder is the appropriate lexicalization, not because the interior of
the cheese cover cannot be seen as a container, but because the physical cover itself
does not control the cheese (cf. the smoke-sentence (25), where the cover does control
the entity within it).
27. It is probably also possible to conceptualize hoek in (30 b) as a set of lines meeting at
an angle and circumscribing a 2DIM area; cf. (39 b).
28. I am not considering the meaning 'the children formed a circular line' here.
29. One possible explanation for the use of in and op in de schop in de maag and de schop
op de borst could be that Dutch speakers do not interpret maag and borst as unbounded
divisions of the human body. Instead, borst could simply be conceptualized as a 2DIM
surface (which, in a COINCIDENCE relation with x, gives rise to op); maag, on the
other hand, could be conceptualized as the 3DIM, vaguely bounded part of the body
that we refer to as the 'gastric region'. On this view, maag would be a medium of the
type exemplified in (43), and hence, in a COINCIDENCE relation with x, trigger the
use of in.
30. As is shown in Cuyckens (1991), a prototype cannot be isolated equally easily in other
SpPs.
31. In contrast to overlapping featural configurations, transformationally linked semantic
information does not necessarily share any common elements. For instance, a 2DIM
surface and a 1DIM line which both trigger off op in a C O I N C I D E N C E relation
with do not share any common elements; rather, they are transformationally
linked in that one can consider a line as a limiting case of a surface.
References
Bennett, David C.
1975
Spatial and temporal uses of English prepositions: An essay in stratificational
semantics. London: Longman.
Brugman, Claudia
1981
Story of over. [M.A. thesis, University of California, Berkeley.]
Clark, Herbert H.
1973
"Space, time, semantics, and the child", in: . E. Moore (ed.), Cognitive
development and the acquisition of language, 65-110. New York: Academic
Press.
Cooper, Gloria S.
1968
A semantic analysis of English locative prepositions. Bolt Beranek and Newman
report, no. 1587.
Coseriu, Eugenio
1964
"Pour une semantique diachronique structurale", Travaux de linguistique et
de litterature 2: 139-186.
70
Hubert Cuyckens
Cuyckens, Hubert
1991
The semantics of spatial prepositions: A cognitive-linguistic exercise. [Ph. D.
diss., University of Antwerp (UIA).]
Fillmore, Charles
1982
"Towards a descriptive framework for spatial deixis", in: R. J. Jarvella and
W. Klein (eds.), Speech, Place, and Action: Studies in deixis and related topics,
31-59. Chichester: John Wiley.
Geckeier, Horst
1971
Zur Wortfelddiskussion. Mnchen: Fink.
Geeraerts, Dirk
1985
Paradigm and Paradox: Explorations into a paradigmatic theory of meaning
and its epistemological background. Leuven: Leuven University Press.
1986
Woordbetekenis: Een overzicht van de lexicale semantiek. Leuven: Acco.
1987
"On necessary and sufficient conditions", Journal of Semantics 5: 275-291.
1988
"Cognitive semantics and the history of lexical semantics", in: B. RudzkaOstyn (ed.), Topics in cognitive linguistics, 647-677. Amsterdam: Benjamins.
1989
Wat er in een woord zit: Facetten van de lexicale semantiek. Leuven: Peeters.
Greimas, A.
1966
Simantique structurale. Paris: Larousse.
Haiman, John
1980
"Dictionaries and encyclopedias", Lingua 50: 329-357.
Hawkins, Bruce W.
1985
The semantics of English spatial prepositions. Duisburg: L. A. U. D. paper,
no. 142.
1988
"The natural category MEDIUM: An alternative to selection restrictions and
similar constructs", in: B. Rudzka-Ostyn (ed.), Topics in cognitive linguistics,
231-270. Amsterdam: Benjamins.
Herskovits, Annette
1986
Language and spatial cognition: An interdisciplinary study of the prepositions
in English. Cambridge: Cambridge University Press.
Katz, Jerrold J.
1972
Semantic theory. New York: Harper and Row.
Katz, Jerrold. J. Jerry A. Fodor
1963
"The structure of a semantic theory", Language 39: 170-210.
Lakoff, George
1982
Categories and cognitive models. Duisburg: L. A. U. D. paper, no. 96.
1987
Women, fire, and dangerous things: What categories reveal about the mind.
Chicago: University of Chicago Press.
1987
Foundations of cognitive grammar. Vol. 1, Theoretical prerequisites. Stanford:
Stanford University Press.
Leech, G. N.
1969
Towards a semantic description of English. London: Longman.
Lyons, John
1968
Theoretical linguistics. Cambridge: Cambridge University Press.
1977
Semantics. Cambridge: Cambridge University Press.
Miller, George A. Philip N. Johnson-Laird
1976
Language and perception. Cambridge: Cambridge University Press.

Pottier, B.
1963
71
Recherches stir I'analyse semantique en linguistique et en traduction mecanique.

Nancy: Publications linguistiques de la Faculte des Lettres et Sciences humaines de Nancy.
Quirk, Randolph Sidney Greenbaum
1973
A university grammar of English. London: Longman.
Rosch, Eleonor
1975
"Cognitive representations of semantic categories", Journal of Experimental
Psychology 104: 192-233.
1978
"Principles of categorization", in: E. Rosch and B. Lloyd (eds.), Cognition
and categorization, 27^18. Hillsdale, N. J.: Lawrence Erlbaum.
Rosch, Eleanor Carolyn B. Mervis
1975
"Family resemblances: Studies in the internal structures of categories", Cognitive Psychology 7: 573-605.
Saussure, F. de
1916
Cours de linguistique generale. Paris: Payot.
Taylor, John R.
1989
Press.
Vandeloise, Claude
1986
L'espace en frangais. Paris: Seuil.
Dividing up physical and mental space into

conceptual categories by means of
English prepositions
Rene Dirven
Introduction
Like any other word category, prepositions structure a subjective, language-specific view of relations in our experiential world. The purpose
of this paper is to show that the structuring of our experience of physical
space by means of prepositions to a large extent determines the languagespecific concepts built up in mental space. More specifically, the paper
aims to analyse the following theses:
(i) Some, if not all, English prepositions are dividing up physical space
in an idiosyncratic, "English" way.
(ii) The basic spatial conceptualisations can be and are projected onto
"mental space", i. e., they may form chains of meaning covering various
conceptual domains such as time, state, area, manner or means, circumstance, cause or reason, etc.
(iii) Since each of the prepositions can follow a more or less similar path
of extensions in meaning, we may, theoretically speaking, find a number
of different concepts of time, state, area, manner, cause etc.
The question therefore is to find out how similar or how different are
the various concepts in mental space, e.g., the various concepts of
circumstance or cause denoted by the different prepositions. Although
one may assume that concepts of cause are fundamentally different from
concepts of manner or area, the question also arises whether the various
concepts denoted by the chain of meanings of one preposition, e.g., at
may not be as closely related in meaning to each other as they are to the
corresponding senses in the domain denoted by another preposition, e. g.,
on. More importantly, one may also expect gaps in the meaning extensions
for some mental domains because of the specific spatial structuring
conventionalised by a specific preposition. These hypotheses will be tested
on a sample of twelve prepositions selected more or less at random: the
three basic space prepositions at, on, in:1 the two "proximity" prepositions
74
Rene Dirven
by and with; the two "path" prepositions through and about, the two
"vertical space" prepositions under and over, and the tree "separation"
or "source" prepositions from, off and out of.
1. Characterisation of spatial conceptualisations

These 12 prepositions and some of the relationships between their spatial
meanings can be brought together as in Figure 1.
This representation of the relationships is, of course, too strong in
some respects. But what it intends to express is the following:
At, on, in are the basic and most general place prepositions. At, as the
most "neutral" place preposition, denotes place as a point of orientation,
disregarding its physical shape: e. g., at the station takes the station as an
orientation point for some trajector, 2 but does not denote whether the
trajector is near, inside or on top of the station area. On denotes physical
contact between trajector and landmark, and therefore necessitates viewing the landmark as one-dimensional space (a line) or two-dimensional
space (a surface). In denotes the enclosure of the trajector in the landmark,
and therefore views the landmark as two- or three-dimensional space (a
surface or a volume).
Similar, but now "separation" spatial configurations hold for the three
source prepositions. From denotes separation from a point of orientation,
e. g., he comes from the station. Off denotes separation from contact with
[1]
[2]
[3]
AT
ON
IN
[6]
[4]
BY
THROUGH
[5]
[7]
ABOUT
WITH
[8]
[91
UNDER
OVER
FROM
[10]
Figure 1. Some relations between prepositions
OFF
[11]
OUT OF
[12]
Dividing up physical and mental space
75
a line or surface (thus a somewhat funny sentence like he comes off the
station could only mean that he had climbed onto the building) and out
of denotes separation from the inside of a landmark as in he comes out
of the station.
Obviously, on and off are the clearest instances of the spatial sense of
all prepositions, which may be a consequence of their basic meaning of
denoting physical contact (or breaking it off) between trajector and
landmark. Although at and in can be used to fulfil purely spatial functions, they do so in a less concrete or less "spatial" way than on, and
therefore they may be more apt to go beyond spatial conceptualisations.
The same holds for the "proximity" prepositions by and with, which
locate the trajector in relation to a point of orientation just like at does,
whereas through and about presuppose a two- or three-dimensional area
just like in.
More specifically by denotes the idea of "connection", either in a static
relationship, e. g., He is sitting by the fire, or in a dynamic relationship,
e. g., This bus passes by the sation. With also denotes both a point of
orientation and the idea of connection, but it subsumes these notions
under the more abstract notion of association and/or accompaniment.
Hence in its spatial sense, with can only denote persons as a landmark,
e. g., John is sitting with Mary.
In contrast with the "point-of-orientation" character of by and with, the
two- or three-dimensional prepositions through and about require the
landmark to be seen as a surface or a volume as in he walks through the
station or he walks about the station (platforms). Thus through structures
space as a tunnel or channel, whereas about denotes spatial movement in
any direction.
The two vertical space prepositions under and over are outsiders in the
general picture drawn here. Their inclusion in the discussion has a dual
purpose: (i) to show that prepositions denoting a negative polarity
this also applies to the "separation" prepositions from, off and out of are
less apt to be extended into all domains of mental space, and (ii) to show
that the positive polarity item over behaves differently from under, even
to such an extent that it can cover some of the domains expressed by
through and especially about.
Consequently, the only real outsider in Figure 1 is under (especially in
view of the position it takes up in the representation). But its position is
somehow justified in that it negates over, which in some of its extensions
adjoins about. Needless to say, there is no link between the adjoining
76
Rene Dirven
Figure 2. Sets of prepositions and relationships
items with and under nor between under and from. The various relationships that do hold between the twelve prepositions chosen are therefore
rather to be interpreted in the sets or groups as represented in Figure 2.
2. Chains of meaning from physical into mental space

The extensions of the meanings of a preposition from physical space via
time into more abstract domains do not occur in any haphazard way but
follow a path of gradually increasing abstractions, whereby the link with
each prior meaning remains obvious and may account for most, if not
all, co-occurrence restrictions between trajector and landmark. Note,
however, that these notions will gradually come to be used in a more
abstract sense, too.
2.1. At extends from an orientation-point in space to one in time, and
further into state, area, manner, circumstance and cause, as illustrated
below:
(a)
(b)
(c)
(d)
point as place: at the station

time-point: at six o'clock
state: at work
area: good at guessing
77
(e) manner (point on scale)
(d) area-
(a) point in space
(c) state
-(b) point in time
( 0 circumstance
(g) cause
Figure 3. Radial network of extensions of at
(e) manner: at full speed

(f) circumstance: at these words (he left)
(g) cause: laugh at, irritation at
The notion of orientation-point obviously links space (a) and timepoint (b); also the state (c) is a metaphorical orientation-point since it
contrasts with other states such as at rest, at sleep, at prayer etc. The
notion of "area" is used in Radden's (1989: 448) sense of "the thematic
context or field within which an event is seen"; thus one can be "good"
within the context or field of "guessing". The fact that at is used with a
number of predicates such as good, bad, clever, adept, an expert etc.
suggests that the skill in question is a point at which the subject of these
predicates is aiming. With the manner expression at full speed (e), at
denotes a point on a scale, which can be contrasted with other points.
Circumstantial at in at these words (f) is a further abstraction of temporal
at, now not with a point on the time continuum but with human actions
as a reference-point. Although it is possible to imply a cause-effect
relationship with circumstantial at in (f), this is not necessary. When this
causal relationship is explicitly given as in (g) laugh at, irritation at,
infuriated at, angry at, surprised at etc., what is expressed is that the
emotional state follows from a cause or that the cause triggers the
emotional state.
We propose the following diagram (Figure 3) to represent these relations.
2.2. On is substantially different from at due to its notion of "contact"
and can have the following extensions:
78
Rene Dirven
(e) means
(d) area
(a) contact with line/surface
(b) time-expanse
(c) state
(f) circumstance
(g) reason
Figure 4. Radial network of the extensions of on
(a) contact with line/surface: on the floor

(b) period of time: on Sunday
(c) state: on display
(d) topic or area: lecture on history
(e) manner/means: dine on snails
(f) circumstance: on arrival (register first)
(g) cause/reason: congratulate sb. on his performance
The notion of spatial contact with a line or surface in (a) is mirrored
on the time-axis in (b) and in the continuing state in (c). The idea of
lecturing (d) presupposes solid contact with one's subject matter. Similarly, snails or whatever other means of subsistence (e) with dine on, live
on etc. form a basis of support.
Consequently, the notion of manner seems to be incompatible with on.
Only the notion of condition, which can be a subcategory of circumstance,
seems to be compatible with on as on her own, on these conditions etc. In
contrast with this concept of a supporting circumstance, the notion of
circumstance with on in (f) does not form a basis of support, but rather
denotes the contact that holds between two successive moments in time
or in an action chain. Finally, the notion of contact seems to exclude the
idea of cause and effect, but the notion of reason is fully compatible with
on, since the reason for the act of congratulation follows from and is
based upon the performance delivered.
2.3. In conceptualises space as an enclosure or volume, and this basic
characteristic also pervades all its extensions:
79
(e) means
(d) area
(a) spatial enclosure
(b) time-span
(c) state as enclosure
(e) manner as state
(f) circumstance as state
(g) cause as state

Figure 5. Radial network of extensions of in
(a)
(b)
(c)
(d)
(f)
(g)
(h)
spatial enclosure: in the station

time-span: in one day; in a week
state as enclosure: in despair, in search of
area: specialise in, rich in coal
manner, means: in a loud voice, in English
circumstance: she nodded in agreement
cause: delight in one's success, revel in
The enclosing space can be physical as in (a) or temporal as in (b): here

it can be a time-span within which one is situated or a later time-span
(seen from the present time-point). The notion of enclosure is extended
to psychological states such as in despair or active states such as in search
of (c). Also the notion of area can be an action field {specialise in) or a
thematic field {rich in). English categorises all kinds of state as enclosing
experiences which may denote manner or means (f), circumstance (g) or
even cause (0; in themselves these expressions merely denote an "enveloping" state, and the further specifications arise in the given contexts.
This may be summarised in the following diagram.
2.4. By highlights the notion of "connection" between two entitities in
physical space or two events or states in mental space.
80
Rene Dirven
(d) area
(lawyer by
profession)
(a) connection in place (house by the bridge)
(b) connection in time

(by day)
(f) circumstance
(by such bad weather)
Figure 6. Radial network of extensions of static by
(a)
(b)
(c)
(d)
(e)
(f)
(g)
connection in space: a house by a bridge', pass by a bridge

connection in time: by day, by six o'clock
area: a lawyer by profession
manner, instrument: travel by bus
circumstance: by accident, by such bad weather
cause/agent: surprised by his appearance, a book by Lakoff
Unlike the three basic prepositions at, on, in, which are mostly static, by
can denote both static {a house by a bridge) or dynamic (pass by a bridge)
relationships. This also applies to time relations: by day denotes a static
relationship, by six o'clock refers to a later moment of time, before which
something is to happen. English does not associate a psychological state
with by (c), which may be due to its strong notion of connection. Also
area is only marginally present in (d). But whenever two entitities, events
or states can be seen as strongly in connection with each other, by is very
productive as with manner, which is usually conceptualised as an instrument (e), or with circumstance (f) and cause or agent (g).
The two diagrams in Figure 6 and 7 are proposed to represent the
extensions of static by and dynamic by, respectively:
2.5. With conceptualises the notions of association and accompaniment,
which are not physical in nature, though they may denote physical
proximity; nevertheless this is not transferrable to time or state:
(a)
(b)
(c)
(d)
(e)
(f)
(g)
proximity in space: John is sitting with Mary

area: deal with a problem
manner, instrument: with great precision, with a key
circumstance: (I cannot do it) with everybody laughing
cause: tremble with fear
81
(e) means/instrument
(a) connection as path

(pass by the bridge)
(b) connection with time-point

(finish by six 'clock)
(f) circumstance
(by accident)
(g) cause/agent
(idestroyed by fire/the enemy)
Figure 7. Radial network of extensions of dynamic by
All the more abstract meanings of with invoke a metaphorisation of the

notion of accompaniment: in the domain of area, the thematic field is
seen as a companion one has to deal with; manner or instrument are
likewise seen as companions to fulfil an assignment; circumstance is a
companion standing by and cause a feeling or emotion leading to visible
effects.
The diagram in Figure 8 represents these extensions:
2.6. Through is, just like about, a path preposition and typically occurs
within dynamic contexts. This basic tendency also prevails in the non(e) manner/instrument as accompaniment
(d) area
(a) proximity/accompaniment in space
(f) circumstance as accompaniment
(g) cause as accompaniment

Figure 8. Radial network of extensions of with
82
Rene Dirven
(a) movement in enclosure
(b) movement in time-span
(e) means as channel
(g) cause as channel

Figure 9. Radial network of extensions of through
spatial usages, which are, however, precisely because of the necessity of

dynamic contexts, more limited than those of by.
(a)
(b)
(e)
(g)
Place as path: walk through the fields

Time as path: go on through the next months
Means: funded through our budget
Cause: killed through accidents
The following diagram can be proposed to represent these extensions:

2.7. About denotes movement in any direction, which also leaves fewer
options for figurative extensions:
(a)
(b)
(d)
(g)
Place: about the playground

Time: about then
Area: think/doubt about
Cause: excited/crazy about
The notion of mental movement is found with verbs denoting speaking

and thinking (d): these activities are seen as movements over a mental
area. The same notion underlies emotional state predicates which denote
emotional states caused by wandering over certain causes (g).
(d) area
(a) movement in all

directions in space
(g) cause
Figure 10. Radial network of extensions of about
(b) approximative time
83
2.8. Over can denote static and dynamic situations and has similar
possibilities to about:
(a)
(b)
(d)
(g)
Place: over the mountain.

Time: over the whole year.
Area: debate over
Cause: argue/fight over
Figure 11 represents these extensions of over:

(d) area
(a) movement to end

of surface
(b) movement in time-span
(g) cause
Figure 11. Radial network of extensions of over
2.9. Under denotes a static position at the negative and of a vertical

dimension, so that it can easily lead to the expression of negative states
or circumstances, too:
(a)
(c)
(f)
(g)
Place: under the bridge

State: under attack, under arrest
Circumstance: under no circumstances, under the premise
Cause: under the thought of
Figure 12 represents these extensions:

(a) lower point in vertical space
(c) state
(f) circumstance
(g) cause
Figure 12. Radial network of extensions of under
84
Rene Dirven
The three "source" or "separation" prepositions are even more strongly

limited in their possibilities for metaphorical extensions of meaning.
2.10. From denotes a point of departure either in physical space, in time,
or in the domain of causation:
(a) Place: He has returned from England.
(b) Time: He has been working from 6 o'clock onwards.
(g) Cause: He died from drugs.
2.11. Off reduced to of in its area or causal meaning, denotes departure
from a surface with which one was in direct contact:
(a) Place: He has come off the ladder.
(d) Area: Talk of the devil (and he is sure to appear.)
(g) Cause: He died of cancer.
2.12. Out of denotes departure from an enclosure, which in a metaphorical sense can be an emotional state causing an effect:
(a) Place: He walked out of the house.
(g) Cause: He killed her out of despair.
In order to summarise the varying possibilities for meaning extensions,
let us first survey the twelve prepositions with respect to their values for
each of the six areas as in Table 1.
Table I. Survey of the meaning extensions of prepositions
at
on
in
by
with
through
about
over
under
from
off
out of
time
state
area
means
manner
circumstance
cause
reason
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
85
Two conclusions that complement each other can be drawn from this
summary and the foregoing analyses:
(i) a preposition that denotes a vaguer or more general location is more
apt to develop metaphorical extensions;
(ii) a preposition that denotes a more concrete location or a specialised
visual location or which has a negative polarity meaning is less apt to
develop metaphorical extensions.
These conclusions do not only apply when larger sets of prepositions
or prepositions within sets are compared, but also when single prepositions are compared:
The first five prepositions {at, on, in, by, with) are vaguer in exact
location than the seven others, and they show more extensions, too (see
Table 1).
Within this set of five prepositions only the first three prepositions
{at, on, in) can form "state" prepositional phrases. In fact, this is not
surprising, if one takes into account the "proximity" meaning of by and
with: a state cannot easily be seen as being near an entitity, but it must
somehow "touch" that entity.
Within the set of the three most productive prepositions at, on, in, on
is more concrete than at and in, since it has, as a major component, the
notion of "physical contact". This feature may explain why on does not
form "active" state predicates like at or in can {at work, in search of): on
just like under only forms "passive" state predicates such as on display
("being displayed") or under arrest ("being arrested"). Furthermore, on
can not form manner phrases, but only expressions of means {diet on
bananas), which is in line with the notion of physical contact. Also the
circumstance meaning of on is a very special one, since it requires two
successive acts touching upon each other {on arrival). The dominating
feature of contact may also explain why the "cause-effect" pair is not
found with on: cause and effect may be a too complex and too abstract
relation to be limited to continuous or even tangent events. But reason
is possible, precisely since one event can be supported (via a logical step)
by some other event.
Also through and about differ slightly in concrete visualisation: through
implies the image of a tunnel or channel and therefore excludes the
notions of state, area, or manner, but only allows "passage-like" concepts
of means or cause; about is much vaguer due to its visual conceptualisation
of "movement in any direction" and allows extension into mental movement required for area {think/speak about) or cause {excited about).
86
Rene Dirven
Over, though only expressing, in its prototypical senses, vertical location, also implies movement and this may be the reason why it has the
same, albeit less productive, possibilities of denoting area and cause. The
negative polarity item under, on the other hand, can only be extended to
notions of passive state (under arrest) and of circumstance (under the
premise).
The three "separation" prepositions are all negative in orientation
and, just like under, allow far fewer extensions into mental space. Only
the most neutral of these three, from, can be used for time. The fact that
all three can be extended into cause suggests that extension into cause,
though a fairly abstract notion, poses a more general problem. In fact,
all these prepositions except on may denote cause, which again requires
an independent explanation, which may be attributable to the notion of
cause rather than to the nature of the prepositions themselves. This will
be the topic of Section 3.5., but it will be discussed in the context of the
differentiated concepts which can be shaped within one domain.
3. Differentiated concepts of place, time, state, area,

manner etc.
Linguistic description finds no difficulty in distinguishing between various
concepts in the domain of place or time. Concepts like point in space or
time-point are familiar ones. So are concepts like line, surface, volume,
enclosure in space or time. The use of different prepositions for the
domains of state, area, manner or means, circumstance, and cause or
reason, suggests that linguistic description cannot but make similar distinctions within these domains, too. I shall therefore now examine the
paradigm of prepositional phrases within one conceptual domain and
investigate how similar or different these expressions are. To facilitate
the comparison, I shall quote the same examples as before, but add more
instances, and group them according to domains. We can leave out place
and time, since here we can assume general agreement, and begin with
state. For the numbering of the prepositions, the order given in Section
1 will be used.
3.1. State
English prepositional predicates denote four different types of state:
(1)
(2)
(3)
(4)
87
at: at
workjrestjsleep)workjplay)prayer/warjlunch
on: on
display/show/sale/hire/trial/guard/duty
in: in despair/sorrow/love/fun/search
of/demand
under: under
arrest/repair/control/fire/pressure
As the examples suggest, these states incorporate important differences.

With at all the prepositional predicates denote a more or less clearly
delimited and usually and "active" state, whereby the subject is usually
human and engages in some form of organised activity; it is the activity
itself which is the point of orientation so that the use of at is fairly
normal. With on (and also with its antonym under) the prepositional
predicates denote "passive" states: things are or can be displayed, shown,
sold, hired etc. Most of these ow-predicates are associated with things,
and only a few with persons: also in this case the "passive" state (on
trial) or the "commissioned" state (on guard, on duty) are still prevalent.
It seems that the on predicates visualise things on a platform for everybody
to see or witness. In this respect the under-predicates are slightly different:
here we find a stronger suggestion of the presence of an agent who
controls (very typical is under control) the situation, /-predicates, finally,
leave room for both passive and/or emotional expressions (in despair/
sorrow/love, in demand), or more active ones (in search of/in a hurry).
This may be seen as a consequence of the "enveloping" sense implied in
the idea of "enclosure" expressed by in: the emotional state, passive state
or active state is viewed as an enclosure in which the subject finds himself.
The four prepositions thus help to conceptualise different forms of
states: at states refer to (parts of) a larger scheduled network of human
activities or states; in states denote feelings, emotions or other individual
situations; on states, on the other hand, denote controlled, often passive
situations, which in adverse conditions are seen as emanating from higher
forces, typically expressed as under states. English thus contrasts at liberty
with in freedom: at collocates with oberty since this contrasts with a
transient state of captivity, whereas in denotes a person's state of being
free; 3 under arrest denotes the control "over" a person, and on trial
denotes somebody's being raised for public investigation.
3.2. Area
Eight of the twelve prepositions can be used to denote area:
(1) at: good/bad/clever/an expert/adept at
(2) on: concentrate/meditate/an expert/lecture/a book/a summit/a report/
a comment on
88
Rene Dirven
(3)
(4)
(5)
(7)
(8)
(11)
in: specialisejrichjlowjpoorjlacking
in
by: a lawyer by profession
with: deallbusylfamiliarlbe
engaged with
about: think/doubt/speak/say
sth./talk/a book about
over: debate/a controversy/a dispute! quarrel I argue over
of: speak / think/remind/know /dream/read of
The point-of-orientation meaning of at is preserved in its transfer f r o m

place to the " a r e a " domain in that this area must be one of practical
skills such as good at playing the piano. In this respect at in area expressions is very similar to at "active" state expressions, where both action
and h u m a n subjects also seem to be m a j o r co-occurrence restrictions.
In contrast with at, the use of on in area expressions presupposes longer
and more p r o f o u n d contact with, preferably, a mental area: thus one can
be "an expert at repairing pianos" in the sense that one practises this
oneself, but be "an expert on (repairing) pianos", which implies knowledge a b o u t the area, which may be the specific field of piano repairing
or the more general field related to pianos in general. This also explains
the co-occurrence of on with such mental activity verbs as concentrate,
lecture, brood, ponder, muse, meditate, etc. The difference between on and
in stands out clearly with the item expert, too: one can be an expert on
the nineteenth century, but for a larger area English prefers in e.g., an
expert in literary history. This notion of area with in is therefore the wider
context within which one is an expert, whereas on designates a more
specific area (of knowledge) and at an area of practical skills. This specific
nature of the enveloping /'-area also accounts for expressions such as
rich/poor/low in: here the notion of richness or scarcity only applies within
a specific field.
The use of with for area evokes the idea of accompaniment, be it
activity as in be engaged with, busy with, knowledge as in familiar with
or mental activity as in deal with. But this sense of with is not very
productive with verbs.
In fact, the four prepositions with verbs denoting mental and/or communicative activity are on, about, over, and of. On and of are, just as in
their spatial meanings, clear antonyms: serious, p r o f o u n d mental activity
requires on: thus lecture on and speak of are almost antipodes. This also
explains that things connected with one's imagination or intuition are
expressed with of. dream of, know of, think of, remind of. The same holds
for things that have to be retrieved f r o m m e m o r y such as remind of,
reminiscent of, oblivious of. The spatial notion of "separation" of the
89
preposition o f ( f ) does not guarantee any in-depth concentration on

things.
As against all these more "specific" conceptualisations of area, the
expression of area by means of about tends to be that of a "dispersive"
or a "discursive" area. This may be based on the fundamental sense of
"movement in all directions", whereby the notion of movement may
explain the co-occurrence restriction to mental activities such as speak,
write, lecture, think, doubt or mental products such as books, stories,
films, articles, papers, studies, songs etc.
Although about and over are similar in some respects, they differ in
one important respect, too: whereas about denotes movement in all
directions, over visualises a two-direction movement, implying the way
from and back to the mental starting-point. It is this two-way sense of
over that shapes the notion of an antagonist area such as in debate,
dispute, quarrel, argue, controversy over.
The four prepositions that cannot be used for the domain of area are
through, under, from and out of The image schema of a tunnel may block
any association with the notion of area and so may the concrete vertical
image schema of under do. Separation prepositions seem by nature less
apt to denote area, although o f ( f ) can be used for the suggestion of a
"distanced" area, an area only vaguely reachable. The question, however,
is whether this of can still be considered as one of the senses of polysemous
ofif)
3.3. Manner, means, instrument

Only six of the twelve prepositions selected can be used to express one
or more aspects of this complex domain:
(1)
(2)
(3)
(4)
(5)
(6)
at: at full speed, at the top of his voice.

on: dine on snacks, drunk on whisky, on foot, on horseback
in: in agreement, in a loud voice, write in ink/pencil
by: by train/bike/car/air/sea
with: with precision/care /passion', with a key /a pen
through: funded through our budget, obtain sth. through the post.
The domain of manner, means and instrument requires prepositions

denoting orientation, contact, enclosure or connection between two entities, situations or concepts.
90
Rene Dirven
The fact that the other six of the twelve prepositions about, over, under,
from, off and out of do not form prepositional phrases of manner, means,
instrument may be due to the vaguer, dispersive meanings of the former
two (about, over), and to the negative polarity of the latter four.
The domain of manner, means and instrument can be seen as a
conceptual continuum, at one end of which we find more abstract conceptualisations for manner and at the other end the more concrete
conceptualisations for instrument; the concept of means, though somewhat closer to instrument stands between the two extremes of manner
and instrument, as shown in Table 2.
Table 2. Distribution of prepositions over manner, means and instrument
MANNER
with
at,
MEANS
INSTRUMENT
with
in,
on, by,
through
The prototypical preposition covering both the concepts of manner

and instrument is with: combined with abstract nouns, it usually expresses
manner, with concrete nouns it expresses an instrument. The other prepositions tend to favour one of the segments of the continuum: at tends
towards the more abstract segment; on and by towards the intermediate
segment of means; in covers the abstract manner segment (nod in agreement) or the intermediate "means" (write in ink), and through conceptualises means as a channel.
The difference between means and instrument stands out most clearly
in the pair write in pencil and write with a pencil: the '-phrase conceptualises the way (manner or means) in which things are done; here the
idea of an enveloping state or substance is clearly present, which also
accounts for the use of pencil as an uncountable noun; the wzfA-phrase
stresses the use of an accompanying instrument and since this is concrete,
pencil is used as a countable noun here.
The fact that on does not favour a manner interpretation is a consequence of the notion of physical contact of spatial on; in its semantic
extension to the manner/means/instrument domain on cannot fully switch
to the fully abstract end of the continuum, although such a manner
interpretation is never fully excluded in expressions such as on foot, on
horseback. But even the other extreme interpretation of instrument cannot
be excluded in these two expressions; still the absence of articles or plurals
in such idiomatic phrases suggests that these are not comparable to freely
91
manipulative instruments, but are rather supporting entities or surfaces

with which the user is in physical contact. This notion of a supporting
means (which also holds for subsistence by means of food and drink) is
different from the "path" expressions denoting means, such as by boatj
bike/airjsea or through our budget etc.; here the notions of "connection"
(by) or "channel" (through) are extended into that of "means", but this
time not a supporting means, but rather a means allowing the passage
of entities. Typically, the concrete entities (boat, bike, bus) or the substances (air, sea) are not seen as fundamentally different since they are
all treated as uncountable nouns.
With through the noun remains countable, however, and therefore such
expressions as through our budget never allow a "manner" interpretation,
which could still be invoked for ^-phrases.
3.4. Circumstance
Five of the six prepositions (i.e., at, on, in, by, with, through) that can
express the domain of manner, means and instrument can also be used
for the domain of circumstance; through is out now, but under takes its
place.
(1) at: at these words (he left)
(2) on: on arrival, on his death, on my return, on receipt, on the condition
that, on the pretext that
(3) in: (he smoked) in silence
(4) by: by accident, (catch) by surprise, by such bad weather
(5) with: (I can't do it) with everybody laughing', with the door wide open,
the bugs can get in
(6) under: under these circumstances, under the premise
As with the former domain, the domain of circumstance requires a more
or less clear "location"; this condition is also met by under.
Circumstance is a conceptual domain that refers to a situation occurring
or holding at the same time as or in a close vicinity of some other
situation; though some link between the two situations is implied, the
nature of this link is not specified or not clearly specified. In fact, there
is not a clear-cut boundary between circumstance and cause or reason,
but, as with most conceptual categories, a fuzzy zone of transition is
found here. Still, the centres of both categories "circumstance" and
"cause" are relatively stable. And even within the domain of circumstance,
92
Rene Dirven
the various prepositions divide up this 'mental space' into quite different
concepts of circumstance.
With at circumstance is viewed as an orientating activity marking the
occurrence of a second activity; as with the other uses of at the notion
of activity prevails here, too. With on two meanings of circumstance
emerge: either the notion of a bordering action (on arrival) which is to
be followed by new action (clearly the notion of contact between the two
actions is present here); or else we have a static circumstance (on the
condition that, on the understanding that), which implies that one situation
is to be conditionally supported by some other fact.
As already suggested in Figures 3 and 4, the extensions of at and on
go from the time concept to that of circumstance so that we can speak
of "temporal circumstance" here. To this we must also add the notion of
"conditional" circumstance.
In denoting circumstance is an extension of state and manner (see
Figure 5). The enveloping metaphor following from the "enclosure" sense
of in does not allow the interpretation of successive events with /-phrases,
but only that of simultaneous situations. In such an "enveloping" circumstance, the link between the two situations is always felt to be stronger
and the borderline with the domains of manner, purpose, cause, or reason
is hard to draw: thus one could interpret she nodded in agreement not
only as manner but also as circumstance, or even as purpose or reason.
The inherent "connection" sense of by makes this preposition almost
into the predestined item to express the neutral domain of circumstance,
especially when no visible link between two events is implied, as in the
expressions by accident, by chance, by mere coincidence. Also other uncontrollable situations such as weather conditions typically require by (by
such bad weather), although here the weather can also be seen as an
enveloping factor and thus allows the use of in: in such a bad snowstorm,
in the rain, in this heat etc.
The "accompaniment" meaning of with, just like the "connection"
sense of by, is a very apt candidate for a meaning extension into the
domain of "circumstance". The difference is that with requires a clearly
delimited situation, which can function as an accompaniment to some
other situations. Consequently, mere indications of time or factors of
luck are not sufficient, but only a fully spelled out situation (e. g., with
the door wide open) can be used for this type of circumstance.
Circumstance denoted by under (under these circumstances) is very
much the same as static circumstance denoted by its antonym on (on the
condition that) in that both tend to express the notion of condition. The
93
difference between on and under in this respect is that on constitutes the

unmarked, neutral case, whereas under is used for the marked case of
adverse situations.
Finally, it should be kept in mind that, although there is no clear
dividing line between the various domains, their centres stand firm. Thus
the expression suffer in translation is unequivocally circumstance, not
cause, although one might, logically speaking, imply a causal factor. Still,
it is not expressed by in here. Expressing such a causal relationship with
suffer requires denoting the source of the process as in suffer from a
tendency to exaggerate. The various types of causal relationship are
discussed in the next section.
3.5. Cause and reason

As already stated before, all the twelve prepositions can denote some
type of cause (or reason):
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
at: laugh at, irritation at, angry at

on: congratulate on, compliment on, pride oneself on
in: delight in, rejoice in, exult/triumph/revel in
by: surprised by, a book by
with: tremble with fear, hair grey with age, pleased with, besides
himself with, blush with pleasure, blind with passion, white with anger
through: killed through accidents
about: excited/crazy/angry/unhappy
about
over: argue over, fight over, hesitant over
under: suffer under a regime
from: die from drugs
of: die of cancer
out of: kill sb. out of despair
The causal meaning of each of these prepositions is strikingly different

and helps to explain the type of verb that can be used in the accompanying
verb phrase.
The three basic prepositions at, on, in are rather restricted in the
expression of causal relations. Thus causal at denotes a cause as a target
at which one aims one's emotions such as laughter, irritation, anger etc.
implying that the target (person or situation) was the cause of the emotion
felt by the subject. Thus cause as target is an extension of the basic spatial
sense of at, but taken in its dynamic sense. The fact that at only takes
94
Rene Dirven
predicates denoting an emotional state or action results from the basic

target sense of dynamic at: since this image schema of target remains
intact, the verb or predicate must be such that it resembles in some
respects the pure target verbs such as aim at, drive at, fly at, get at etc.
(see Dirven 1989: 548).
Causal on is also closely linked to the image schema of support
associated with spatial on: thus one situation is felt to be the ground on
which the other rests one congratulates, or compliments somebody basing
oneself on the achievements of that person. This image schema also
explains why on does not denote cause proper but only reason: the fact
that one situation rests on another does not mean that the second triggers
off the first, at least not in English. But in the two related languages
German and Dutch, the closest historical items are an and aan, which
are very generally used for the expression of causal relations. But an and
aan have only kept the contact notion of on, not the support notion,
which in German and Dutch is expressed by auf and op, respectively.
Causal in is like all other meanings of in based on the spatial image
schema of enclosure and the mental schema of an enveloping state: e. g.,
in delight in hard manual work the manual work is the state enveloping
and causing the emotional state of delight. 4 Consequently, the predicates
that can be used with causal in must also denote an emotional state which
is caused and kept going by the 'enveloping' state denoted by the in
phrase. The list of such predicates is rather limited and only contains
predicates which denote a strongly felt emotion such as delight, rejoice,
exult, triumph, revel etc.
Summarizing these three types of cause, one can make the generalisation that the three basic prepositions only allow a type of cause or reason
which determines human or animate behaviour and which is restricted
to emotions.
A similar, though less strong, restriction holds in the case of causal
with: this is mostly used with predicates denoting perceptible animate
behaviour such as tremble /grin/blush/beside oneself/blind/ white with (the
latter three in their metaphorical senses) or perceptible states or processes:
hair grey with age; echo with the clash of cutlery, etc. These co-occurrence
restrictions for causal with follow from the basic sense of accompaniment:
the cause is seen as an accompanying factor of some behaviour or state
and triggering it off. For this reason, causal with strongly contrasts with
the causes denoted by the three basic prepositions which do not require
perceptible situations such as delight, rejoice etc.
95
By and through denote a cause which is seen as a path and therefore

have the potential of denoting a more general type of cause. In actual
fact, this is only reserved for by in English, through being too much
concretely visual in its tunnel image schema. 5 In the sentence Many people
are killed through accidents, the through phrase denotes the notion of
being involved in an accident with the effect of being killed. This visual
configuration is absent in killed by, which therefore is not compatible
with circumstantial causes such as accidents. Here English only expresses
the circumstance by means of in, e. g., killed in an accident, but not the
agentive cause, e. g., * killed by an accident. Consequently, the notion of
causal by requires an independently acting cause, often bearing "primary
responsibility though not volition" as Lakoff (1987: 66) puts it. This is
nicely illustrated in a sentence like The tense situation was resolved by the
entrance of the corporal, where the by phrase denotes a cause, but not a
willing agent. But it would take us too far to go into the distinction
between cause and agent here.
Although about and over are also path prepositions, they do not denote
a linear path as by and through, but a scattered path in all directions
{about) or a path in its back and forth directions {over). This makes these
two prepositions apt to be used for emotional causes. Causal about and
over are very strongly related to the 'area' meanings of these two prepositions (see Figures 10 and 11). The extension of the area meaning found
with verbs of cognition and communication such as know, doubt, wonder,
write, speak, talk goes via a transition zone such as worry about to purely
emotional states such as denoted by
excited/crazy/angryj{un)happy/confident about etc. In all these predicates the subject is human and seems
to be moving "emotionally" about the cause of his emotion.
Since the movement associated with over is less hectic and implies a
two-way or a repeated movement, causal over is limited to reciprocal
verbs such as fight, argue, etc. or to predicates implying repeated thought
movements such as hesitant over. The image schema underlying expressions such as worked up over may be that of a vertical dimension only
whereby the state of being "worked up" is caused by the event situated
below it. Causal under denotes a cause working from above so that the
state in which the subject is results from the weight of that cause: this
image schema is present in suffer under a regime, her heart sinks under
the thought of, stagger under the impact of, agonise under somebody's
sarcasm, etc.
Of the twelve prepositions under discussion, the three "separation"
prepositions from, of and out of, are least restricted in denoting cause.
96
Rene Dirven
Since the differences between a remote cause (die from drugs), an immediate cause (die of cancer) and a non-focussed cause as state (kill
somebody out of despair) have already been analysed in great detail by
Radden (1985: 88 ff.), we can here dispose of the question of why the
image schema of "separation from a source" is so apt to become metaphorised into the expression of cause.
Conclusion
The above analysis of the radial networks of meanings of English prepositions and of the differentiated senses within one domain of meaning
has amply shown that the uses and possibilities of the prepositions to
express other than spatial meanings is not an arbritary or random hocuspocus, but a highly motivated system. The most striking phenomenon is
the relationship between the way physical space is divided up in English
and the way mental space is structured. It is due to the very specific way
prepositions are geared to denote relationships in the domain of spatial
experience that English has made differentiations in the domains of
mental experience such as area, manner or cause.
More generally, we can conclude that languages, even the most related
ones, have conceptualised the links between spatial and mental domains
of experience in slightly or markedly different ways. The basis of it all is
the conceptualisation of physical space. The structurings that have taken
place here also determine the later extensions of these spatial concepts
into the mental domains.
Notes
1. The terminology used here and in the following paragraphs to characterise the meanings
of three English prepositions is the same as that used in Dirven (1989) and is accounted
for there.
2. The notions trajector and landmark are used in the specific sense given them by Langacker
(1987).
3. Dictionaries are not always very consistent in pointing out this difference, but vaguely
imply it. Thus libery is defined in The Shorter Oxford English Dictionary as "exemption
or release from captivity, bondage, or slavery", freedom is defined as "exemption or
release from slavery or imprisonment, personal liberty".
4. Cobuild English Language Dictionary contrasts delight in hard manual work with delight
at the prospect of leaving home; this use of at is necessary since a prospect is not
97
conceivable as an enveloping state; the -phrase is intermediate between circumstance

(at these words he left) and cause and hence a typical example of the continuum scale
between circumstance and cause.
5. Note that other languages have the equivalent of through for cause/agent, e. g., German
durch (but only for inanimate causes/agents) and Dutch door for all types of animate
and inanimate causes/agents.
References
Dirven, Rene
1989
"Space prepositions" in: R. Dirven (ed.), 519-550.
Dirven, Rene (ed.)
1989
A user's grammar of English: Word, sentence, text, interaction. Frankfurt: R
Lang.
Lakoff, George
1987
Langacker, Ronald
1987
Foundations of cognitive grammar. Vol. I: Theoretical prerequisites. Stanford,
CA: Stanford University Press.
Paprotte, W. T. Rene Dirven (eds.)
1985
The Ubiquity of Metaphor, Metaphors in language and thought. Amsterdam:
J. Benjamins.
Radden, Gnter
1985
"Spatial metaphors underlying prepositions of causality", in: W. J. Paprotte
- R. Dirven, (eds.), 177-207.
1989 a
"Semantic roles" in: R. Dirven (ed.), 4 2 1 ^ 7 2 .
1989 b
"Figurative use of prepositions", in: R. Dirven (ed.), 551-576.
On the grammar of lexical and non-lexical

prepositions in English
Gisa Rauh
1. Introduction
Discrepancies and inconsistencies have almost become an integral characteristic of the description of English prepositions. Thus, for example,
Poutsma (1926: 761) claims that in English prepositions, in addition to
their normal uses, are used "as the analytical equivalents of what in
Present or Old English, or some cognate languages is, or at least may be,
synthetically expressed by inflection". This applies in particular to of, to,
for and by. In the same spirit Curme (1935: 29) introduces the notion
"inflectional particles" for these cases, and Deutschbein (1917: 278) even
talks of "analytical case", referring to to as dative and of as genitive.
Strongly rejecting such views, Jespersen (1924: 186), on the other hand,
emphasizes that the so-called "analytical cases" must not be separated
from other prepositional groups: "It is much sounder to regard them as
what they really are, prepositional groups." Thus Jespersen demands that
all prepositions be treated as parts of prepositional groups or phrases;
however, the proposals made by Fries (1952) and subsequent structuralists
(e.g., Hockett 1958; Gleason 1955; 1963; Stageberg 1965) point in the
opposite direction. They regard all prepositions as function or structure
words, thus grouping them together with affixes signaling gender or case,
with conjunctions, negation particles, auxiliaries and articles. For these
authors prepositions are distinguished from nouns, verbs, adjectives and
adverbs. Their exclusion of prepositions from the latter group is justified
by pointing out their lack of lexical content (e. g., Fries 1952: 106) or by
claiming that they form a closed set (Stageberg 1965 [1977]: 239).1 In a
similar way Fillmore (1968; 1969) describes prepositions in the framework
of Case Grammar as variants of inflectional units. Although he distinguishes three types of preposition those with lexical content, those
dependent on verbs and those expressing underlying case relations (cf.
Fillmore 1969: 367) syntactically they are all described in the same
way, i. e., as markers on noun phrases. According to Fillmore (1969: 367)
100
Gisa Rauh
the notion prepositional phrase can be done away with altogether since
prepositions do not represent the heads of constructions:
Thus we see that the distinction between noun phrase and preposition
phrase is no longer necessary. This is all to the good of course, since
preposition phrase has always been a terminological nuisance. We would
really like all constituents labeled X-Phrase to be constituents having X's
as their heads.
The opposite view, and thus one similar to that of Jespersen, is taken
by Jackendoff (1973, 1977, 1983) who, like others, criticizes Fillmore's
position:
It is often claimed that prepositions are not a lexical category, but rather
that they are simply case markers on noun phrases, possibly even inserted
by transformations. This approach is institutionalized in Fillmore's Case
Grammar (Fillmore 1968); Postal (1971) trivializes the role of prepositions
still further ... Though diachronically the reduction of prepositions to case
markers is undeniable, it is not clear that such a process plays a role in
synchronic grammar. (Jackendoff 1977: 80)
He, in turn, describes prepositions as elements of an autonomous lexical

category on a par with the categories noun, verb and adjective. In general,
this view is adopted in the framework of X-bar-syntax (cf. Emonds 1976;
van Riemsdijk 1978; Stowell 1983; Chomsky 1986 a, 1986 b; Baker 1988).
Emonds (1985), however, again distinguishes prepositions from the lexical
categories noun, verb, and adjective since, according to his view, they are
"a grammatical rather than a lexical category" (1985: 14), even though
they represent the heads of constructions.
In this paper it will be demonstrated on empirical grounds that a
uniform description of various uses of prepositions, as suggested by
Jespersen, Fillmore or Jackendoff, is not justified. It will be shown that,
based on syntactic evidence and with some consideration of semantic
properties prepositional variants can be distinguished which can be
divided, to start with, into two classes, lexical and non-lexical.
"Lexical" here is intended to mean that these items are elements of a
lexical category characterized by syntactic, semantic and phonological
properties analogous to those of other lexical categories. Lexical prepositions have their own autonomous lexical entries. Non-lexical prepositions, on the other hand, do not have autonomous lexical entries of their
own. They are not autonomous lexical items. They are the result of
various grammaticalization processes with syntactic and semantic consequences, as characterized, for example, by Meillet (1912) and, more
recently, by Lehmann (1985; 1986/87). Since various processes are involved, the set of non-lexical prepositions is heterogeneous. It is the aim
Lexical and non-lexical prepositions
101
of this paper to set up criteria which provide an explicit basis for the
identification of lexical and non-lexical prepositions and, furthermore,
for a characterization of types of non-lexical variants.
In what follows, the properties of lexical prepositions will be indicated.
These are comparable to those of elements of other lexical categories and
may be considered as conditions on lexical heads. Then two types of
variant of non-lexical preposition are discussed whose specific properties
are compared to and contrasted with those of lexical prepositions. A
brief summary concludes the presentation.
2. Lexical prepositions
Elements of lexical categories are both syntactically and semantically the
heads of constructions. As syntactic heads they determine syntactic properties of constructions, e. g., syntactic projection properties, the formation
of category-specific proforms, the potential for coordination or movement, and Case properties. As semantic heads they determine semantic
properties of constructions such as, for example, semantic "projection
properties", which include the selection and theta properties of internal
arguments, modification properties related to elements in adjunct position
and specifying properties related to elements in specifier position. They
determine semantic properties of proforms as well as conditions on
coordination. As semantic heads, elements of lexical categories also determine their argument structure, rendering possible via their external
arguments a connexion with the heads of other constructions which
cannot be provided on purely syntactic grounds. Finally, as semantic
heads, elements of lexical categories are characterized by inherent semantic properties which distinguish them from other elements of the same
category in a more or less significant way.
The properties indicated hold for elements of lexical categories in
general. They also hold for a subset of prepositions.
2.1.
Syntactic properties of lexical prepositions
2.1.1. Projection properties

According to the assumptions of the Theory of Principles and Parameters
(cf. Chomsky 1981, 1986 a, 1986 b) unmarked phrase structures conform
102
Gisa Rauh
to the conditions of X-bar-theory. Thus, categories which are not the

head of a construction are projections thereof. The facts for English are
expressed by the following schemata (Chomsky 1986 a: 3):
(1)
a. X' X"*
b. X" - "* X'
In (1 a) X refers to the head of a construction and X"* to its complements,

"*" indicating that their number may be > 0. In (1 b) X"* refers to
the specifier of X. According to Chomsky, the schemata are valid for the
lexical categories noun [ + N, V], verb [N, +V] adjective [ + N , +V]
and preposition [ , V] as well as for the non-lexical categories COMP
and INFL.
Studies by Jackendoff (1973, 1977), van Riemsdijk (1978) and Emonds
(1972, 1976, 1985) have provided empirical evidence for the fact that
prepositions in English exhibit projection properties of the kind indicated.
It is emphasized here that this holds only for a subset of prepositions or
their uses, i. e., for lexical prepositions. Just like verbs, these allow various
XP-categories as their complements ( = "C-selection", cf. Chomsky
1986 b: 86), whereby again like verbs individual prepositions may
differ with respect to their C-selection properties. The following examples
demonstrate that prepositions, in addition to allowing various XP-complements, may be intransitive:
(2)
a. John stayed [PP in [ NP the house]],

b. John stayed [PP in].
(2')
a. John [VP cooked [ NP the dinner]],

b. John [VP cooked],
(3)
a. John went [PP into [ NP the house]],

b. *John went [PP into].
(3')
a. John [VP salted [ NP the dinner]],

b. *John [VP salted],
(4)
a. I didn't see him [PP before [ NP the movie]].

b. I didn't see him [PP before [CP the movie began]].
c. I didn't see him [PP before],
(4')
a. Bill [VP knew [NP the answer]].

b. Bill [VP knew [CP the answer was correct]].
c. Bill [VP knew].
(5)
a. Bill stayed [PP until [ NP dawn]].

b. Bill stayed [PP until [CP the day dawned]].
c. *Bill stayed [PP until].
(5')
a. Bill [VP recognized [ NP his friend]].

b. Bill [VP recognized [CP that it was his friend]].
c. *Bill [VP recognized].
(6)
a. John came [PP from [ NP the station]].

b. John came [PP from [PP behind the curtain]].
(6')
a. The pencil [Vp pierced [ NP the paper]].

b. The pencil [VP pierced [PP through the paper]].
103
The description of [P + XP]-constructions as PPs, as in (2) to (6'), with

XP realized as NP, PP, CP or 0, is based on the observation that they
exhibit analogous syntactic behaviour. Thus, for example, each of the
PPs may be fronted, inducing what is called "comma intonation" (cf.
Emonds 1985: 249):
(7)
a. Before the movie, I didn't see him.

b. Before the movie began, I didn't see him.
c. Before, I didn't see him.
The fronting of directional PPs is accompanied by NP-V-Inversion, as

Jackendoff (1973: 347) points out:
(8)
a. From the station came John.

b. From behind the curtain came John.
c. In came John.
PPs with various XP-complements may occur in cleft sentences:

(9)
a.
b.
c.
d.
It
It
It
It
was
was
was
was
[PP before [ NP the movie]] that I saw John.

[PP before [CP the movie began]] that I saw John.
[PP before] that I saw John.
[PP from [PP behind the curtain]] that John came.
They can also be coordinated:

(10)
a. [PP Before [CP the movie began]] and [PP until your arrival] it
was very boring.
b. He rushed [PP in] and [PP down [ NP the stairs]].
c. John came [PP from [PP behind the curtain]] and [PP off [ NP the
chest]].
104
Gisa Rauh
In addition, prepositions with various XP-complements may be replaced

by proforms which can only replace PPs, not VPs, APs or NPs:
(11)
a. Bill refused [PP before I did] and John

before me
before
refused [PP then] too.
b. The first session [VP lasted an hour] and
[VP so] did the second session too.
*then
c. The first meeting was [AP late] and [AP so]
*then
was the second too.
d. Bill loves [ NP the summer] and John loves [ NP it] too.
*then
(12)
a. Bill stayed [PP up the hill] and John

up
stayed [PP there] too.
b. Bill [VP lives in London] and [PP so] does
* there
John too.
c. The kitchen is [AP ten feet long] and [AP so] is the living room.
*there
d. Bill hit [ NP the fence] and John hit [ NP it] too.
* there
As an additional piece of evidence for the adequacy of classifying the

various [P + XP]-constructions as PPs, Jackendoff (1973: 347) and
Emonds (1976: 174; 1985: 257 ff.) cite the fact that in each case right may
occur as a specifier:
(13)
a.
b.
c.
d.
John
John
John
John
stayed right in the house.

stayed right in.
came right from behind the curtain.
arrived right before Mary did.
It is significant that right is not allowed in the context of VP, NP or AP:

(14)
a. *Mary [VP right invited John].

b. *Mary is [ AP right beautiful],
c. *Mary is [ NP right a beauty].
105
Reference to right as a specifier introduces properties of beyond the

domain of complements, i. e., to properties above the P1 level. According
to the Chomskyan schemata presented in (1), lexical categories have two
levels of projection, the specifier position in English being defined as the
daughter of X 2 ( = Xmax) and the left sister of X1. In (1 b) the number of
possible specifiers is determined as > 0. Thus, iteration of specifiers is
allowed, but no principled distinction between the various specifiers and
the head is indicated. As Jackendoff s analyses (1977) demonstrate, however, with respect to , V and A at least, this approach is inadequate.
Jackendoff, who assumes Xmax = X3, distinguishes for these categories
specifiers in X 3 as sisters of X2 as well as specifiers in X2 as sisters of X1.
As X3-specifiers he identifies articles like the, a, these, those, etc. for N,
degree terms like so, too, as, etc. for A and the subject of a sentence for
V. As X2-specifiers he identifies quantifiers like many, few, several, etc.,
measure phrases, attributes and postmodifying phrases for N, adverbials
for V and measure phrases for A and P. X3-specifiers differ from X2specifiers in that they close X max -phrases at the periphery. The position
can be filled only once, whereas X2-specifiers may iterate.
Although Jackendoff assumes X m = X 3 for as well as for , , V
and others, he does not provide examples of P3-specifiers. However, he
does not explicitly exclude them either. This is done by Fukui and Speas
(1986), not only for but also for A, and V. These authors make a
distinction between functional categories like INFL, DET and COMP
and the lexical categories , , V and P. They identify X2 as the maximal
projection for functional categories and X1 for lexical categories. X1 is
assumed to allow recursion. Thus Jackendoff s X2-specifiers and some of
his X3-specifiers correspond to daughter and sister of X1 in this model.
Emonds (1985) also assumes recursion of X1. In contrast to Fukui and
Speas, however, he identifies specifiers of XP (X = , , V, P) which
close the maximal projection (Emonds 1985: 18 ff.). According to
Emonds, specifiers are represented by grammatical formatives which are
not heads of constructions, like DETERMINER (this, that, these, those,
the, a, each, every, etc.) as SP(N), INTENSIFIER (very, so, as, more,
most, too, enough, etc.) as SP(A), AUXILIARY (will, would, can, could,
may, might, etc.) as SP(V) as well as the prepositional specifiers right,
clear and straight. In addition, Emonds identifies as specifiers measure
phrases, which occur especially in the context of A and P. Unlike Jackendoff, Emonds thus does not distinguish specifiers on various levels for
, V and A. With respect to P, his solution is de facto the same as
Jackendoff s.
106
Gisa Rauh
Somewhat different are the suggestions made by Radford (1988). He,

too, provides examples for the recursion of X^categories and identifies
the position X'-daughter-and-left-sister-of-X 1 as that of premodifying
elements including e. g. measure phrases. He calls this position, which
corresponds to that of postmodifying adjuncts, "attribute position" and
distinguishes it from the specifier position, defined as daughter-of-X max and-sister-of-X 1 , which closes a maximal projection. In addition to determiners for N, aspect for V and determiners for A (e.g., that, so,
completely, etc.) Radford identifies right as a specifier of R
The examples demonstrate that premodifying elements of have not
been described uniformly in the literature. Further examples of these
diverging approaches could be considered (e.g., van Riemsdijk 1978;
Bierwisch 1988). For the present line of argument it is not necessary to
decide whether elements like right occupy the same position as measure
phrases like, for example, two feet in two feet above the door, that is,
whether they are specifiers or attributes (in Radford's terms). What counts
is simply the fact that prepositions allow premodifying as well as postmodifying elements and thus project beyond the domain of complements.
Syntactically these elements may be either grammatical formatives (in the
sense of Emonds) like right, measure-NPs, or PPs, as the following
examples illustrate:
(15)
The store is [PP right [P- across the street]].
(16)
Bill lives [PP two miles [P< down the road]].
(17)
Bill arrived [PP two hours [P< before Mary]].
(18)
a. Bill went [PP down [P- to his friends]].

b. Bill went [PP [P< [P' down] [PP to his friends]]].2
(19)
He stayed [PP [P< [P< at his sister's] [PP near Brighton]]].
Thus, with respect to projection properties, prepositions exhibit the same

behavior as elements of the lexical categories V, or A. 3
2.1.2. Case properties
In addition to syntactic projection properties, which determine the construction of phrases and thus also make predictions concerning movement, substitution and coordination, prepositions also exhibit Case properties. Pronominal complements as in (20) and (21) provide evidence for
the fact that prepositions are Case assigners:
(20)
Mary got the book from him.
(21)
Bill bought a book for her.
107
In the Case Theory of the Theory of Principles and Parameters a distinction is made between syntactic structural and inherent lexical Case.
The latter is related to theta roles (Chomsky 1986 b: 153). It is postulated
that an element assigns inherent Case to an N P if it theta marks this
NP. Since prepositions theta mark their complements, Case assigned by
prepositions is considered inherent Case. Assignment of inherent Case is
postulated for , and A, explicitly restricting it to the Genitive for
and A, realized by either 's or of in the context of and only by of in
the case of A. The assignment of structural Case is postulated for V and
INFL and is independent of theta marking. We will see later that there
is more to be said about the assignment of Case by prepositions (Section
4.). At this point it suffices to have pointed out that prepositions do
assign Case, which in English, if morphologically realized, is identical to
the Case assigned by verbs. Therefore, van Riemsdijk and Williams (1986:
229) use the term "accusative" to capture both instances.
2.2.
Semantic properties of lexical prepositions
2.2.1. Semantic "projection properties"

Lexical heads of constructions exhibit not only properties of syntactic
heads, but also of semantic heads. What is important here is the fact that
heads determine or are involved in cooccurrence restrictions which include
the whole domain of the syntactic projection and thus exceed the domain
indicated by Chomsky (1965: 95ff.), which refers to objects, i.e., complements, and subjects of verbs and adjectives only. (22) gives, for elements of the category V, examples of such selection properties related to
complements and determined by semantic properties of the head:
(22)
a. Bill read the book.

*the boy.
her thoughts.
b. The lesson lasted an hour.
*a door.
c. The bug crawled over the wall.
*above
d. The bug flew over the wall.
above
108
Gisa Rauh
In (22 a) the semantics of read requires the complement to express something readable. The adequacy of using her thoughts in this context demonstrates that selection properties are not to be understood as restrictions
requiring a constituent to inherently provide these properties but rather
as transfer features in the sense of Weinreich (1972: 96 ff.) which are
provided by the verb and read into the feature matrix of the complement.
Thus, due to the selection properties of read, thoughts is interpreted here
as something readable. The semantics of last in (22 b) requires a temporally interpretable complement. Other complements are inadequate. In
contrast to fly, the semantics of crawl expresses contact with a surface.
Since above is marked as expressing distance rather than contact between
two entities, it is inadequate in the context of crawl, but adequate in the
context of fly. Since over is unmarked with respect to distance, it may
head a PP in the context of either verb, as is shown in (22 c) and (22 d).
Similar selection properties may be observed for lexical prepositions:
(23)
a. Bill stayed above the creek.

*an hour,
average.
b. Bill arrived at the station.
five o'clock.
c. The hadn't heard of it since last week.
*the house.
London.
King George.
d. Chaos reigned through the house.
across
e. The pencil is on the desk.
in
(23 a) shows that above does not take a complement which is inherently
marked as temporal. It does, however, take complements which allow a
spatial or a scalar interpretation. The preposition at, on the other hand,
allows spatially and temporally interpretable complements, as is demonstrated by (23 b). This is a first indication that above and at are
characterized by selection properties of two kinds, which means that there
are lexical variants. We will return to this point later. The examples under
(23 c) illustrate that since has temporal selection properties and takes only
complements which are interpretable temporally. 4 London and King
George are interpreted accordingly, paraphrased perhaps as since they
had been in London or since the times of King George. Examples (23 d)
109
and (23 e) illustrate that as a result of different selection properties of the

head in question, the same complement is interpreted in different ways.
Thus through in (23 d) induces a three-dimensional conceptualization of
a house, whereas across induces a two-dimensional one. And on in (23 e)
causes the selection of a two-dimensional part of a desk, unlike in, which
causes the selection of a three-dimensional one.
The following examples demonstrate that elements of the categories
N, A and V also determine cooccurrence restrictions between heads and
pre- or postmodifying elements, i. e., elements in specifier/attribute and/
or adjunct position: 5
(24)
a. many of the men

*milk
b. much of the milk
*men
c. a group of men
*milk
d. a cup of milk
*men
e. two feet of rope
*milk
(25)
a.
so J *pregnant
b. utterly incompetent
*tall
c. six feet tall
*pregnant
*
incompetent
d. six months pregnant
*tall
incompetent
(26)
Bill was drawing a circle.

*Bill was knowing the answer.
(27)
Bill liked the woman in the green shirt.

*idea
(28)
Mary is pregnant with a boy.

*tall
110
Gisa Rauh
(29)
Bill believed in the stork for several years.

*reached the top
(30)
Bill broke the vase using a hammer.

*read the book
As the following examples illustrate, lexical prepositions exhibit the same

cooccurrence properties:
(31)
a. two meters above the door

behind the door
*before noon
* above average
*on the door
b. three minutes before noon
*above the door
* above average
c. some degrees above average
*above the door
*before noon
(32)
a. Bill went past the house along the river.

*until noon.
b. Bill arrived at five o'clock sharp.
*the station
c. Bill stood behind the door near the wall.
*from the wall.
d. Bill arrived at five o'clock in the morning.
*at the station.
e. Bill arrived at the station near London.
*at five o'clock.
In (31 a) the measure phrase two meters can only specify a spatial preposition and, since this measure phrase expresses a distance a, only one
which itself expresses a > 0. Since on expresses contact and thus = 0,
two meters is inadequate as a specifier in this context. Examples (31 b)
and (31 c) illustrate that a temporal specifier can specify only a temporal
preposition and a scalar specifier only a scalar one. The aftove-examples
in (31 a) and (31 c) also illustrate that this preposition is polysemous
between two interpretations, one spatial and one scalar. Thus, the potential for different specifiers supports the assumption presented earlier that
two variants may be identified. As with polysemous expressions in general,
111
the semantic variation is evident only in context. If, however, a constituent

in the projection is marked for one of the potential variants of a preposition, this variant is the one selected and the remaining constituents of
the projection are interpreted accordingly. This condition is illustrated
by the examples under (32 b), (32 d) and (32 e), which demonstrate that
at is polysemous and has a spatial and a temporal variant. In (32 b) the
complement five o'clock is marked for the feature [temporal]; thus the
temporal variant of at is selected and sharp interpreted accordingly. The
complement the station, however, selects the spatial variant of at. Since
sharp cannot be given a spatial interpretation, the sentence is ill-formed.
Similarly, in (32 d) only the temporal PP in the morning is possible as an
adjunct to temporal at, not the spatial PP at the station. If the latter is
analyzed as an adjunct to V, the expression is well-formed. The situation
is analogous in (32 e) where, because of the station, the spatial variant of
at is selected. In a similar way, in (32 a) the until-PP is incompatible with
the head past if the former is analyzed as an adjunct to the latter. In
(32 c) the from-PP is inadequate as an adjunct to behind since it expresses
a source, which cannot serve as a modification of behind.
These examples should be sufficient to demonstrate that lexical heads
exhibit what are here called semantic "projection properties".
2.2.2. Semantic substitution properties
A related and yet somewhat different property of lexical heads may be
seen in the fact that they semantically determine the syntactic proform
of their maximal projection. This is illustrated by the nominal proforms
in (33) and (34):6
(33)
the boy - he
*she
(34)
the girl she

*he
In a similar way, prepositional proforms are marked for either space or

time:
(35)
a. He stayed in London.
there.
*then.
b. Where did he stay? In London.
*When
112
(36)
Gisa Rauh
a. He came in the morning.

then.
* there.
b. When did he come? In the morning.
*Where
For causal PPs including what is traditionally classified as a causal

clause introduced by a subordinating conjunction a specific whproform may be identified:
(37)
Why did he do it? Because of the rain.

Because it was raining.
In addition, therefore may be considered a causal prepositional proform,

showing by its morphological complexity that causal prepositions are
always derived and never basic:
(38)
Because of the rain he did it.

Therefore he did it.
The proform thus, which may also replace [P + NP)-sequences and which
is inherently marked for "manner" as is illustrated in (39),
(39)
He spoke in a loud voice.

thus.
cannot be considered a prepositional proform since as a deictic, especially

text-deictic expression, it is predominantly semantically marked and may
replace various units of a text, including syntactically unspecified units:
(40)
a.
b.
c.
d.
Thus spoke the chief.

When thus gentle, Mary was the nicest girl in the world.
The legend has thus much of foundation.
The new machine will work twice as fast, thus greatly reducing
costs.
The same holds for the wh-form how, which represents a semantically
rather than syntactically specified proform and may replace various
syntactic categories:
(41)
a. How did he leave? In a hurry.

Quickly.
Convinced.
Cursing.
113
b. How tall is he? - Six feet.

- Very tall.
It seems that the set of possible, semantically marked prepositional
proforms is relatively small and is by far exceeded by the set of meanings
which are attributed to prepositions (cf., e.g., Radden 1982; 1989),
including the spatial, temporal and causal ones identified in traditional
grammar.
2.2.3. Argument structure and theta properties
A further important semantic property of lexical heads is that they
determine an argument structure and a theta grid. 7 Thus, for example,
the verb put provides three arguments, assigning them the theta roles
AGENT, T H E M E and GOAL respectively. What renders argument structure and theta grid so important is the fact that these semantic properties
are mapped onto syntactic ones, providing an interface between syntax
and semantics. Since Williams (1981 a), a distinction has been made
between an external argument and (n > 0) internal ones; the internal
ones being represented by the sisters of the head, the external one being
located outside the maximal projection. In the case of verbs the external
argument is represented by the subject. Thus, the external role of put,
the A G E N T role, is assigned to the subject. T H E M E and G O A L are the
internal roles assigned to the complements, which represent the internal
arguments. In addition to external and internal roles, a referential role is
identified in Theta Theory. Thus, Williams (1981 a: 86) introduces the
role "R" in the context of nouns, which is relevant for referential uses of
NPs. Higginbotham (1985), following Davidson (1967), introduces the
referential event role "E" for verbs of all kinds, not restricting this role
to event verbs as Davidson did. It is postulated that with respect to nouns
the referential role is identical to the external one whereas this does not
hold for verbs.
All lexical prepositions determine a two-place argument structure and
a corresponding theta grid. It is this property which gives them the
relational character which was recognized and emphasized in traditional
grammar, and expressed, for example by Curme (1935: 87) in his definition
of prepositions in the following way: "A preposition is a word that
indicates a relation between the noun or pronoun it governs and another
word, which may be a verb, an adjective, or another noun or pronoun."
The internal argument is always represented by the complement. According to traditional grammarians, including Curme, this complement
114
Gisa Rauh
is always realized as an NP. In Section 2.1.1. it was claimed that it could

also be a CP or a PP. The internal argument is assigned the internal role
under government, a process described by Higginbotham (1985: 559 ff.)
as "theta marking". This role, which behaves like a free variable, is bound
by the complement and thus saturated. If a preposition is intransitive,
the internal argument is implicit. In this case the internal role is bound
by contextual information.
The external argument of prepositions, which is located outside their
maximal projection, is identified according to the position or function
the PP assumes. If the PP is a complement or an adjunct in a construction,
then its external argument is lexically represented by the head of the
construction. If the PP assumes the function of the predicate, it is the
subject. The dependencies may be demonstrated more precisely by considering the assignment, the "discharging" or saturation (cf. Higginbotham 1985: 561) of theta roles. PPs may occur as predicates in socalled "small clauses":
(42)
a. [sc Bill in London] is unexpected,

b. Bill hates [ s c meetings on Sundays].
In cases like these, the external role of is assigned under government

by theta marking as in the case of complements and saturated by
the subject. In sentences like the following,
(43)
a. Bill is in London.
b. The session is in the evening.
the PPs are not predicates but complements of the verb be, which in these
examples is not the copula but a lexical verb which determines an
argument structure of its own, just like stay or take place. The referential,
non-predicative character of such PPs is clearly shown by the fact that
they can be replaced by deictic proforms (cf. Jackendoff 1983: 49):
(44)
a. Bill is there.
b. The session is then.
Sentences like (45),

(45)
Bill is in pain.
in which the sequence [P + NP] is used predicatively, will be considered
at a later stage (Section 3.). Anticipating, I claim that in this case the
preposition is not lexical.
115
If a PP assumes the function of an adjunct, its external role is identified

with the refential role of the head of the construction. It is this latter role
which in turn has to be saturated. In this respect PPs behave like APs,
which are described by Higginbotham (1985: 564). (46) and (47) are
examples in which the external roles of and A respectively are identified
with the referential role R of the nominal head of the construction:
(46)
the day before Sunday

the trip on Sunday
the book on the table
(47)
the early morning

the high board
the blonde girl
In the examples under (48) the external role of is identified with the
referential role of the verbal head:
(48)
a. Bill met Mary in London.

on Friday,
b. Bill slept for hours.
in the afternoon.
If the PP is the complement of a construction, the external role of

is again identified with the referential role of the head, in (49) with that
of a verbal head:
(49)
a. The session lasted until five o'clock.

b. Bill threw the ball behind the tree.
c. Bill jumped on a chair.
Sentence (49 c) is ambiguous between two readings: the PP could equally

be an adjunct to V. The difference between the two readings is reflected
in the way theta assignment works. A PP as a complement, but not as
an adjunct, is assigned a theta role by the verb in addition to theta
identification, and is thus connected with the verb in two ways. Haumann
(1991: 148 ff.) argues convincingly that this explains the often observed
fact that the coherence between verb and complement is much stronger
than that between verb and adjunct, prohibiting, for example, certain
types of movement. The similarity between PP-complements and PPadjuncts, on the other hand, is explained by the fact that in both cases
the external role of PP is identified with the referential role of V.
116
Gisa Rauh
A comparison of the connexion between PP-complements or PP-adjuncts and the head of the construction illustrates how important is the
task performed by those lexical items which may be the heads of adjuncts,
including P. The Licensing Principle formulated in Chomsky (1986 b:
93 ff.) requires that each element in a well-formed structure must be
licensed in one of a few possible ways. Thus, for example, complements
are licensed by thematic roles which they receive from the head of the
construction, predicates in that they require a subject. Adjuncts are not
governed by the head of the construction in which they are included and
thus cannot receive a theta role from it. They are not required by any
constituent and in turn do not require any. The only possible way of
licensing adjuncts is via an unsaturated external role which their heads
have and which needs to be saturated. This is achieved by theta identification with the referential role of the head. Thus, the external role of
its head provides the only connexion between an adjunct and the construction of which it is an integral part. It follows necessarily that lexical
items which may be heads of adjuncts must have an external role which
is not saturated in their own projections. We shall consider some consequences of this requirement later (Section 3.).
With respect to adjunct-PPs a further remark is in place. Since adjuncts
are not governed by the heads of constructions containing them, they
cannot be subject to the selection properties identified in Section 2.2.1.
However, restrictions may be defined on the basis of external arguments.
Thus it was claimed that prepositions express relations, the two semantic
arguments representing the entities related. Given this prerequisite, a
natural consequence is the condition that, with respect to ontological
type, the arguments must not differ. It is, in other words, inadequate to
relate a spatial entity with a temporal one by means of a preposition. If
the internal argument of a preposition is determined by lexical selection
properties, then automatically restrictions are imposed on the external
argument too: the external role of the preposition of an adjunct-PP can
be identified with the referential role of a given head only if the latter
provides an external argument for the preposition which is not ontologically distinct from its internal one. This condition explains the wellformedness or ill-formedness of the examples presented in 2.2.1. to the
extent that it is related to the semantic correspondence between the two
arguments of a preposition in adjunct-PPs and, of course, also in complement-PPs.
The discussion of adjunct-PPs and adjuncts in general necessarily leads
to a conclusion which will be relevant in what follows. If it is the case
117
that adjuncts are licensed only by identifying their external role with the
referential role of the head to which they are adjuncts, then all lexical
elements which are heads and which take adjuncts must provide a referential role. Since prepositions, as was illustrated in Section 2.1.1., are
heads and take adjuncts, it follows that they must provide a referential
role, just like nouns and verbs. This would explain the property of
referentiality of PPs emphasized by Jackendoff (1983). Jackendoff bases
his claim on the observation that PPs can be replaced by "pragmatic
anaphors", i.e., deictic expressions (cf. Jackendoff 1983: 49): 8
(50)
a. Your coat is here [pointing] and your hat is there [pointing].

b. He went thataway [pointing].
The following sentence provides evidence for temporal PPs:

(51)
Push the button ... now!
It follows that a referential role of depending on the semantics of

should be specified as either spatial, temporal or something else.
Assuming, then, that lexical prepositions in addition to their external
and internal roles, provide a referential one, two questions arise: 1. What
is the relation between these roles? and 2. How is the referential role
saturated? With respect to verbs it is assumed that the external and the
referential role are distinct. The referential role is saturated by theta
binding via existential quantification. For nouns it is assumed that the
external and the referential roles are identical. Saturation is achieved by
theta binding either by means of the specifier or by existential quantification (cf. Higginbotham 1985: 559 f.). The uses of PPs which clearly
exhibit referential properties suggest that the situation here is neither like
that for V nor like that for N. The fact that P1 can be replaced by a
deictic proform and the fact that intransitive prepositions can be used
referentially provide evidence for the claim that in the case of prepositions
the internal role is identical with the referential one. Accordingly, in the
context of transitive prepositions the referential role is saturated via theta
marking and in the context of intransitive prepositions by contextual
information. Since the referential role is identical with the internal role,
it is unnecessary to assume several semantically specified referential roles,
as indicated above, because an interpretation of the internal argument as
a spatial or temporal unit is guaranteed by the selectional properties of
the preposition, for example SPATIAL (x) or T E M P O R A L (x) (cf.
Section 2.2.4.).
118
Gisa Rauh
This section has looked at argument structure and the theta grid,
considering how many arguments and theta roles are determined by
lexical prepositions and how the roles are saturated. The specific semantic
content of the theta roles is part of the specific semantic properties which
characterize lexical prepositions as autonomous lexical items. These properties are considered in the following section.
2.2.4. Specific semantic properties
By definition, lexical heads exhibit inherent semantic properties. On the
one hand, elements of the same syntactic category are distinguished with
respect to semantic properties, on the other hand, shared properties allow
the identification of semantic classes. Thus, for example, the nouns man
and woman differ from each other in that the former is marked by the
property MALE (x) and the latter by FEMALE (x). Since both, however,
exhibit the properties ADULT (x) and HUMAN (x), they can be assigned
to the semantic classes of adults ( = ADULT (x)), humans ( = HUMAN
(x)) as well as human adults ( = ADULT (x) & HUMAN (x)).
In contrast to statements about prepositions which classify them as
structure or function words with only structural meaning (e.g., Fries
1952: 106; Hockett 1958: 153; Quirk 1962 [1968]: 79), numerous contributions to the semantics of prepositions have demonstrated that they
exhibit specific lexical meaning. 9 Again it should be noted that this holds
for lexical prepositions. Irrespective of which theoretical position is preferred at present approaches to the semantics of prepositions are
characterized by their diversity rather than by unity, as can be seen from
the contributions to this volume in any of the approaches questions
of the following kind will have to be central, assuming that expressing
relations is a category-specific characteristic of prepositions:
1. What are the properties of the arguments?
2. Which roles are assigned to the arguments?
3. What are the properties of the relation?
In connexion with the first question, the relation to a domain is of
primary importance, i.e., whether the units related are characterized as
SPATIAL (x) or TEMPORAL (x), for example. What is important here
is the characterization of the internal argument since, as was pointed out
before, the external argument may not differ and thus its properties are
derived automatically. If a preposition selects a spatial unit as its internal
argument, as do in, on, behind, near, above, etc., for example, then it
119
belongs to the semantic class of spatial prepositions. If it selects a temporal

one, like, for example, since or until, then it belongs to the semantic class
of temporal prepositions. It may be the case that the specific meaning of
a preposition also defines further properties of its potential arguments.
It is interesting to note that definitions of this kind seem to be restricted
to the internal argument, whose syntactic representation is governed by
P. Thus it seems that semantic properties are assigned under government,
as are Case and theta roles. Semantic properties of the kind indicated
include, for example, in the context of SPATIAL (x) 3DIM (x), as in the
case of in or through, or 2DIM (x), for example on or across, which define
the internal argument as a three-dimensional or two-dimensional spatial
unit. These two properties are responsible for the different interpretations
constructed for the sentences (23 d) and (23 e), presented in Section 2.2.1.
and repeated here as (52 a) and (52 b):
(52)
a. Chaos reigned through the house.

across
b. The pencil is on the desk,
in
With respect to the second question, it is assumed here as suggested

in Rauh (1988) that the set of roles assigned by prepositions is
restricted. The external argument always receives the role THEME,
whereas the internal argument is assigned either LOCATION, in the
context of on, in, at, by, near, etc., SOURCE, in the context of from,
since, o f f , etc., PATH, in the context of along, across, through, during,
etc., or GOAL, in the context of to, toward(s), for, until, etc. The
preposition with assigns its complement the role ACCOMPANIMENT.
That lexical prepositions do, in fact, assign roles themselves and do not
simply express roles assigned by verbs as is often assumed is evident
in those cases where the roles assigned by a verb and a preposition in
the former's domain differ, as in the following examples:
(53)
a. Bill lives along the road.

b. Bill threw the ball on the roof.
In (53 a), live assigns its PP-complement the role LOCATION, along its
NP-complement the role PATH. This expresses that the place where Bill
lives is situated along the path provided by the road. In (53 b), throw
assigns its PP-complement the role GOAL, on its NP-complement the
role LOCATION. As a result, the goal of the throwing is interpreted as
identical to the place expressed by the complement of P: the roof. More
120
Gisa Rauh
frequent than examples like (53) are those where the roles assigned by
verbs and prepositions in their domain are identical, as in (54 a) and
(54 b):
(54)
a. Bill walked along the road,

b. Bill lives on a hill.
This can lead to the inadequate conclusion that prepositions express

rather than assign roles, as is assumed, for example, in the framework of
Case Grammar.
It should be noted that THEME, LOCATION, SOURCE, PATH and
GOAL are considered here as roles independent of any specific semantic
domain. If these roles are regarded as properties assigned by prepositions
to their arguments just like domain-specific selection properties, e. g.,
SPATIAL (x) or TEMPORAL (x), then by conjoining both types of
property domain-specific roles may be defined. Thus, for example, LOCATION (x) & SPATIAL (x) defines a spatial localization, LOCATION
(x) & TEMPORAL (x) defines a temporal localization, SOURCE (x) &
TEMPORAL (x) defines a temporal starting point or area, etc. The
notions LOCATION, SOURCE, PATH and GOAL signal that the relations they express were originally relevant for the spatial domain and
later became transferred to other domains, a development that may be
observed for many prepositions diachronically 10 and in language acquisition.11
With respect to the third question ("What are the properties of the
relation?"), various criteria are relevant. For spatial prepositions these
include, for example, relative distance, such as DIST (x, y) in the case of
above or PROX (x, y) in the case of near or by and CONTACT (x, y) in
the case of on, verticality, such as ABOVE (x, y) for over and above and
BELOW (x, y) for under and below, as well as originally deictically
determined horizontality dependent on the perspective of the observer,
which may be transposed to intrinsically structured objects, such as IN
FRONT OF (x, y) or BEHIND (x, y). Purely topological criteria, expressed, for example, by INCLUDE (y, x) or EXCLUDE (y, x) are also
relevant. With respect to temporal prepositions, a relevant criterion for
the kind of relation is the position within a linear sequence, i. e., either
BEFORE (x, y) or AFTER (x, y).12
Thus, semantic properties of the kind presented may, in fact, be identified for lexical prepositions. We shall not go into the important and yet
unsolved problem of how to represent the lexical meaning of lexical
prepositions. For the present purpose it suffices to have pointed out that
121
the semantics of prepositions always includes three different aspects of

meaning, one related to the arguments, one related to the roles and,
finally, one related to the relation.
2.3.
When is a preposition lexical?
2.3.1. Conditions
The syntactic and semantic properties of lexical prepositions identified in
Sections 2.1. and 2.2. provide criteria on the basis of which lexical and
non-lexical prepositions can be distinguished. The criteria may be interpreted as conditions on lexical autonomy in the following way:
(55)
a.
b.
c.
d.
e.
f.
g.
h.
Conditions on the lexical autonomy of prepositions:

A preposition is lexically autonomous iff it exhibits the following properties:
syntactic projection properties, i. e., complements (n > 0) are
defined with respect to their categorial realization ( = C-selection, cf. Chomsky 1986 b: 86), adjuncts and attributes and/or
specifiers are allowed;
determination of a syntactic proform for P 1 ;
category-specific movement and coordination properties;
Case properties;
semantic projection properties, i. e., is involved in semantic
well-formedness conditions within P max ;
semantic determination of the syntactic proform;
argument structure and theta grid;
definition of properties of the arguments, of theta roles ( = Sselection, cf. Chomsky 1986 b: 86 f.) and of the relation between
the arguments.
In the following sections it will be demonstrated that non-lexical

prepositions fail to fulfill some of these conditions in a significant way.
Before turning to non-lexical prepositions, however, examples of lexical
prepositions will be presented.
2.3.2. Examples
Lexical prepositions first of all include the primary spatial ones. This is
not surprising since, as language historians have repeatedly observed,
they have developed from spatial adverbials and thus from elements of
122
Gisa Rauh
a lexical category (cf. e.g., Whitney 1989 [1969]: 414ff.; Brugmann 1911:
758 ff.; Paul 1920: 3 f.; Behaghel 1924: 29 ff. or Wackernagel 1924: 165 ff.).
Thus, prepositions like in, on or at all exhibit the properties listed above,
though only when used spatially. This may be the case when represents
the head of a complement or adjunct to , V or P. Deadjectival spatial
prepositions, like near or round, also exhibit these properties. Next is
interesting because its adjectival origin is obviously responsible for the
fact that it is incapable of assigning Case and thus requires a preposition,
to, to assign Case to the NP-complement. Except for this Case property,
next exhibits all the properties listed. Denominal spatial prepositions,
mostly phrasal with a nominal element, e.g., in front of, on top of, or
developments of such a construction and now merged like beside, across
or aboard, also exhibit these properties. Interesting here are examples like
in front of or on top of, which represent transitive prepositions to which
there is an intransitive equivalent, as is illustrated by the following
examples:
(56)
a. Go in front of me!
b. Go in front!
(57)
a. Put the book on top of the others!

b. Put the book on top!
It is evident in examples like these that of, as was suggested with respect
to to in the context of next, compensates for the inability of the [ + N]category in this case to assign Case. Thus, only the first two
elements of on top of and in front of represent the lexical preposition,
which because of its deviant Case properties differs from other
lexical prepositions. Similar observations hold for the denominal form
ahead, merged from the sequence [P + NP]:
(58)
a. Go ahead of me!
b. Go ahead!
At a later stage it will be demonstrated that examples like these provide

further relevant insights (cf. Section 4.).
It is a well-known fact that numerous prepositions of spatial origin
may be used non-spatially. Only a few forms, however, may be identified
as autonomous lexical prepositions which fulfill the conditions listed
above. The following examples illustrate that temporally used spatial
prepositions are a case in point:
(59)
123
a. Bill arrived on Friday in the afternoon at about five o'clock,

b. Bill arrived then.
In (59 a) on is the head of a PP with a temporal complement modified

by a temporal i'w-PP in adjunct position, which in turn is modified by a
temporal -. Thus, the in-PP and the a/-PP illustrate that the o-PP
and the /-PP take temporal adjuncts and thus must provide a referential
role. The prepositions on, in and at differ with respect to the arguments
they select. Whereas on requires a time span of one day as its internal
argument, in selects a time of day like, for example, morning or afternoon,
a year, or a season, and at selects a point in time. In each case the
preposition selects a time span. In each case something is located temporally, i.e. the role LOCATION is assigned to the internal argument.
As is demonstrated by (59 b), on determines a temporally marked proform. Thus these prepositions exhibit lexical properties which differ from
those of their spatial equivalents. It is assumed here that this is due to a
reanalysis of lexical properties. In the case of on this view is supported
by the fact that temporal on takes a YP-complement, which is not possible
in the context of spatial on:
(60)
On hearing the news Bill left the room.
It is also the case that spatial, but not temporal in and on may be used
intransitively. Thus, lexical variants are to be identified.
It is easy to see how variants like the above may develop. If it is
assumed that primary spatial prepositions exhibit selection properties on
the basis of which they select a spatial unit as their internal argument,
then it is possible to imagine that deviant marked sequences can be
generated by inserting non-spatial complements, thus violating the selection properties and thus inducing a metaphorical interpretation. As Kittay
(1981; 1987: 214 ff.) has demonstrated, one of the relevant principles of
metaphor is the transfer of structural or organizational properties from
one semantic domain to another. It is then obvious why at rather than
in was selected to express a point in time, and in rather than at to express
the time spans of a time of day, a year or a season. The examples also
illustrate that properties defined with respect to spatial units, like 3DIM
(x) or 2DIM (x), are selected and transformed in a domain-specific way
when transferred to another semantic domain. Whereas area and space
are relevant units with respect to the spatial domain, duration is relevant
with respect to the temporal domain. While concepts like ABOVE and
BEHIND, R I G H T and LEFT yield structural properties for the spatial
124
Gisa Rauh
domain, of these only IN F R O N T and B E H I N D are selected and transformed to BEFORE and A F T E R to express relations in the temporal
domain. The three dimensions of space contrast with the one-dimensionality of time, which allows only linear sequencing of its units.
There are numerous examples of temporal lexical variants of originally
spatial prepositions, including derived forms like before ("bi" ( = by,
about) + "foran" ( = from the front)), after ( " a f ' ( = L. ab) + "ter"
( = farther off)); or until ("on" ( = up to) + "till" ( = goal)). Of these,
before used as a spatial preposition is rare in Modern English, and after
as well as until have completely lost their spatial properties. 13 Changes
like these confirm the assumption presented earlier that reanalysis of
lexical properties takes place. That these forms constitute lexical heads
even in those cases where they take a CP-complement and thus
according to traditional views represent subordinating conjunctions is
evident from the fact that they can be modified by temporal modifiers in
either specifier/attribute or adjunct position:
(61)
a. Before you came, at about five o'clock, they ...

b. Two hours before you came, they ...
(62)
a. After you had arrived, at about five o'clock, they ...

b. Soon after you had arrived, they ...
(63)
They stayed right until sunrise.

the sun rose.
In additon, theta roles are assigned: the role T H E M E (x) is identified

with the role of the matrix verb, while before and after theta mark
their complements with the role LOC (x), and until its argument with
GOAL (x). Tests related to the other conditions also yield positive results.
Obviously, the development from a marked metaphorical use of a form
to a lexical variant characterized by reanalyzed lexical properties takes a
long time, as is documented by observable intermediate stages. It should
also be noted that this process may develop in two divergent directions,
which will be discussed in more detail in Sections 3. and 5.
In addition to temporal uses of spatial prepositions, there are numerous
others, which, in part, have reached a high degree of lexicalization.
Interesting cases are those where obviously the metaphorical use has been
lexicalized. It is characteristic of these forms that the properties of spatial
prepositions are reinterpreted rather than reanalyzed. Since interpretation
is a process related to the conceptual level, these variants will here be
called "conceptual" variants as opposed to "lexical" variants. These
125
include, for example, scalar and sequential uses of spatial prepositions,

as illustrated in the following sentences:
(64)
a. The temperature is two degree above zero.

below
b. The temperature is near
zero.
close to
c. Bill is far above average.
(65)
a. In the alphabet a is two letters before c.

b. John comes two persons after me.
right
before
The spatial prepositions in these examples provide properties which

specify the kind of relation, i. e., verticality and relative distance in (64),
localization on the front-back-axis in (65), and thus provide structural
principles according to which other domains may be structured. They do
not define properties of their arguments, i. e., they do not provide a
domain-specific definition as, for example since or on do in (66) with
respect to their respective internal arguments:
(66)
a. I haven't seen him since London,

b. He arrived on Sunday.
*the summer.
In (67) and (68):

(67)
Paris is above London.
(68)
Paris comes after London.
the property with respect to which Paris and London are ordered on a
scale or in a sequence is not determined. These sentences only express
that they are ordered in these ways. This use of prepositions is characterized by the fact that though they may be modified, this again may be
done only by way of metaphor, i.e., by using spatial expressions or
expressions which may modify spatial prepositions. Identification of the
relevant domain is possible only via nominal or verbal expressions, or
on the basis of further linguistic or extra-linguistic context. It is also
significant that there are no proforms corresponding to spatial and
temporal prepositional proforms. Thus it would be inadequate to talk of
scalar or sequential prepositions in the same way as of spatial and
temporal ones. These variants do not represent lexical prepositions, which
126
Gisa Rauh
require a lexical entry of their own, but rather lexicalized conceptual

variants dependent on the lexical spatial variant.
In addition to spatial and temporal prepositions, some causal and
modal forms like because of, on account of, and by means of, in accordance
with, exhibit all the properties of lexical prepositions: they define their
internal argument as a causal or modal unit, they assign theta roles, they
take specifiers/attributes and adjuncts which are domain-specific, and
they determine domain-specific proforms:
(69)
a. Bill did it just

because of Mary on behalf of her father.
especially
b. Bill did it especially on account of his failures on behalf of his
friend.
c. Why did he do it? Because of Mary.
On account of his failures.
(70)
a. Bill did it almost by means of will power by virtue of his

firmness.
b. Bill did it especially in accordance with
almost
his instructions by means of his will power.
c. How did he do it? By means of will power.
In accordance with his instructions.
In addition, causal and modal prepositions exhibit a variety of C-selection

properties, including the selection of various categories, as is shown in
the following examples:
(71)
a. Bill did it in order [ CP PRO to win].

*[to [ NP John]],
b. Bill did it because [of [ NP Mary]].
(*of) [Cp Mary wanted it].
(72)
a. Bill did it by means of [ NP his strength],

(*of) *[CP Mary wanted it],
b. Bill did it in accordance with [ NP his decisions],
(*with) *[CP he decided].
Examples like (71 b) illustrate what has already been pointed out with
respect to deadjectival and denominal spatial prepositions, i. e., that the
preposition preceding the NP-complement compensates for the lack of
Case assigning properties of the complex preposition, which seems to be
inherited from its nominal origin. Since lexical NPs, but not CPs, must
127
be Case marked, only in the former context is a prepositional form like

of, to, with, etc. needed. These latter are, in general, considered constituents of phrasal prepositions (cf., e.g., Sweet 1892: 134 f.; Poutsma 1926:
715ff.; Kruisinga 1932: 382 f.; Quirk and Mulholland 1964). We will
return to this subject at a later stage (Section 4.).
All of the causal and modal prepositions discussed so far exhibit all
the properties of lexical prepositions. There are others which do not.
Some do not determine a proform:
(73)
a. Bill did it especially in regard of his exam.

*Why did he do it? In regard of his exam.
In regard of what did he do it? In regard of his exam,
b. Bill did it especially in contrast with his friends.
*How did he do it? In contrast with his friends.
In contrast with whom did he do it? In contrast with his
friends.
Others neither determine a proform nor take specifiers/attributes or

adjuncts:
(74)
Bill did it *especially by virtue of his strength

* almost by way of fighting
*by way of fighting.
*How did he do it? By virtue of his strength.
By virtue of what did he do it? By virtue of his strength.
They share this scale of deviation from complete lexicality with numerous
phrasal prepositions, which are rather specific semantically and cannot
be included in general classes such as causal or modal. To these belong
at the cost of, at the risk of, at the desire of, in favour of, in preference to,
in relation to, and many others. Expressions like these are relevant in
connexion with the non-lexical prepositions discussed in the next section.
3. Prepositions in fixed phrases

A large number of prepositions occur in phrases which constitute syntactically invariant syntagmas. In addition, these constructions often
exhibit fixed semantic and/or morphological properties. In the examples
under (75) the noun expresses a normal constitutive aspect of a person
or thing, in (76) a measure unit:
128
Gisa Rauh
(75)
out
out
out
out
out
out
of
of
of
of
of
of
shape
form
order
character
tune
temper
(76)
by the pound
by the meter
by the dozen
In the examples under (77) and (78) the noun is always deadjectival, in
those under (79) and (80) deverbal:
(77)
on the quiet
on the sly
on the cheap
(78)
in
in
in
in
the
the
the
the
(79)
on
on
on
on
on
on
the
the
the
the
the
the
(80)
in the know
in the swim
clear
dark
raw
open
beat
hop
mend
move
run
trot
In constructions like these the prepositions are not the head of the
construction, neither syntactically, nor semantically. They take neither
specifiers, which are possible in the context of lexical prepositions, nor
attributes or adjuncts:
(81)
a.
b.
c.
d.
e.
Bill
Bill
Bill
Bill
Bill
is (*right) out of shape (*near twenty pounds).

bought the coffee (*right) by the pound (*in a bag).
had a whisky (*right) on the quiet (*on the sly).
is (*right) on the move (*near London).
is (*right) in the swim (*in the Thames).
The accompanying NP is fixed, i. e., it cannot be altered:
(82)
a.
b.
c.
d.
e.
129
Bill is out of (*good) shape.

*Bill bought the coffee by two pounds.
Bill had a whisky on the (*stealthy) quiet.
Bill is on the (*long) move (*to London).
*Bill is in good know of the events.
The preposition does not determine a proform, neither syntactically nor

semantically. In some cases how may occur, a form which may be substituted for predicative APs and their adverbial variants, as (84) illustrates:
(83)
a. How did he buy his coffee? By the pound.

b. How did he drink his whisky? On the sly.
c. How is Bill? *Out of shape.
*On the move.
*In the know.
(84)
a. How is she? Nice.

b. How did he do it? Nicely.
c. *How girl is she? Nice.
We shall see later on that this is significant. Whether prepositions in these

constructions have Case properties can only be inferred via analogy, not
proved. In English, Case marking is visible only in the case of nominal
proforms like him, her, me, etc., which are not allowed here. Since,
however, of as part of the owr-o/-constructions is assumed to be a Case
assigner (cf. Section 4.), it may be concluded by analogy that in the
constructions containing primary prepositions the latter also assign Case
to the following NP.
The argument structure and theta grid in these constructions cannot
be identified in the same way as it can for lexical prepositions. The fact
that these constructions never occur as complements in referential positions but rather either as predicates or adjuncts, and the fact that they
do not determine deictically used proforms and do not take specifiers/
attributes or adjuncts indicate that no referential role is involved here.
On the contrary, it may be concluded that it is the lack of a referential
role which explains the observed syntactic properties. An external role
and, accordingly, an external argument must be present, however, since,
as was pointed out in 2.2.3., adjuncts are licensed only by their external
roles. If the expressions are considered as semantically marked, metaphorically interpretable constructions, the external role may be identified
as the THEME. On the same basis, then, an internal non-referential role
may be identified. In the examples presented it is the role LOCATION.
130
Gisa Rauh
Again on the same basis, it may be stated that the preposition determines
the kind of relation. From the set of properties of the spatial lexical
prepositions those properties are selected which fit the metaphorical use,
and they are interpreted accordingly. In the case of in this is the relation
INCLUDE (y, x), for out of EXCLUDE (y, x) and in the case of on
SUPPORT (y, x). According to the claims and partial results of recent
research on metaphors, which also deals with prepositions, cognitive
principles can be identified which determine restrictions on the possibility
and order of the transfer of spatial properties to non-spatial semantic
domains and thus allow predictions about possible arguments of certain
prepositions. 14 Applied to the present case, this would imply, provided
the principles of order and transfer are known, that the prepositions in
these contexts behave as regularly as the conceptual variants discussed
in 2.3.2. Of these it was claimed that, as lexicalized metaphorical uses
dependent on spatial lexical prepositions, they are represented in the
lexical entry for the latter. However, comparing the two types in question
leads to the conclusion that the syntactic deviation of the constructions
under investigation here also has semantic effects. At the same time it
becomes evident that metaphorical use of language may induce different
processes of reanalysis, which first result in lexical and finally also
syntactic changes (cf. Section 5.).
Comparing conceptual variants like scalar and sequential uses of spatial
prepositions with the uses of spatial prepositions considered here reveals
a significant difference. Whereas the former remain heads of constructions, with the same syntactic properties as their spatial sources and with
reinterpreted semantic properties, this is not the case with respect to the
latter. As was demonstrated above, they are definitely not syntactic heads
of the constructions. Strangely enough, however, in these constructions
the accompanying noun is not the head either, as is illustrated by the
inadequacy of replacing it by a proform:
(85)
a. *Bill had a drink what? - On the quiet.

*On what did he have a drink? On the quiet,
b. *Bill is what? In the know.
*In what is Bill? In the know.
Thus, these constructions are exocentric and therefore marked, i. e., they
do not conform to the principles of X-bar-theory. The fact that some of
these constructions may be replaced by proforms which may also replace
predicative adjectives and their adverbial variants, as was pointed out
131
before, as well as the fact that these constructions take the adjectival or
adverbial specifier so rather than any prepositional specifier,
(86)
a. She is so nice.
She did it so nicely.
b. Why did you do it so on the quiet?
c. Why was she left so in the dark?
d. He is always so on the run.
e. He is so in the swim nowadays.
the fact that they coordinate with APs or AdvPs rather than with genuine
PPs,
(87)
a. He had a drink on the quiet and quickly.

*in the house.
b. Bill is in the clear and innocent.
*at his sister's.
c. Bill is on the move and happy.
*in London.
d. Now Mary is in the know and nervous.
*in London.
the fact that, like APs and AdvPs, they may function as adjuncts or
predicates, and not as complements, like genuine PPs, but sometimes
even as attributes,
(88)
a. He is an out of work teacher.

b. He is good at after dinner speeches.
c. He watched an on-line operation.
and, finally, the fact that in Modern English there are many expressions
classified as adjectives and/or adverbs which have developed out of
[ +NP] constructions,
(89)
a. A: awake, aghast, alive, alert, akin

b. Adv: again, apart, aloud, aboard, ashore, apace
c. A, Adv: alike, ajar, aloof, alone, amiss
all this provides evidence for the claim that what is involved here is a
kind of word-formation process on a phrasal basis, which can be explained on semantic as well as syntactic grounds.
Obviously there is, first of all, a communicational need to express
semantically differentiated properties, circumstances or states as modifications with respect to a given referent for which the language does not
132
Gisa Rauh
provide adjectives/adverbs and which are essentially characterized by the

semantics of given nouns. NPs, however, cannot function syntactically
as adjuncts since there is no way to license them in this position. They
do not provide an unsaturated external role which could be identified
with the referential role of the head of the construction since the role in
question is already saturated inside the N P by theta binding. Semantically,
isolated nouns or NPs cannot express properties, circumstances or states
as modifications of given referents since, again, this capacity is dependent
on a kind of external argument that is lacking for nouns. The relevant
syntactic properties that nominal heads need to function as adjuncts and
the semantic properties needed to express properties, circumstances or
states as modifications can be provided by prepositions which, for their
part, lack the capacity to express the required semantic differentiation.
Thus the [P + NP]-constructions under consideration can be viewed as
a symbiosis of elements of two categories, each one providing properties
which are inherited by the new product thus created. The inheriting,
syncategorematic, syntactically marked construction X receives the external role and the external argument as well as relational properties from
P, and semantic properties from N. 1 5 The internal role of is, as in
compounds or derivations, saturated by the N P inside the construction.
The construction X is, however, syntactically as well as morphologically
marked. It is neither a phrase conforming to syntactic principles, nor a
word conforming to morphological ones. Therefore, expressions like those
presented represent one lexical unit. Of these, units exhibiting the same
patterns, sharing syntactic, semantic and morphological properties, are
related by lexical rules. The symbiotic relation between and NP leads
to both together being considered a unique semantic unit and interpreted
accordingly. Deriving the original metaphorical interpretation requires a
conscious act of separating the set of features and redistributing them to
their respective sources.
Since the properties of such Xs disregarding their position relative
to nominal heads correspond to those of adjectives or adverbs, they
may be used like the latter. A recategorization of these marked structures
in the direction of adjectives or adverbs, reducing their degree of markedness and finally reintegrating them into the grammar, begins when they
start taking adjectival or adverbial specifiers. It may be considered completed when the two constituents merge to form one phonological unit
and thus one word, representing an element of one category and exhibiting
the properties of this category. On the basis of the properties pointed out
here, this type of use of prepositions might be called "adjectivizing" or
133
"adverbializing", independently of whether the final step of lexicalization

has taken place, since prepositions in this context allow NPs to be used
as adjectives or adverbs.
4. Case prepositions
A small set of prepositions occurs in [P + NP] constructions strictly
subcategorized by verbs, adjectives or nouns, being lexically required
and in this sense governed by these categories:
(90)
a. Bill [VP believes in science],

b. Bill is [ AP good at tennis],
c. Bill is [Np an expert on instruments].
In general, the sequence [P + NP] is described as a PP-complement (e.g.,

Chomsky 1965: 101 f.; Akmajian, Steele and Wasow 1977; Bresnan 1978:
18 f.; 'Grady 1985; Radford 1988: 344 ff.), with the selection of a specific
preposition noted as an idiosyncratic lexical property in the following
way:
(91)
a. believe. V, [_ [PP in NP]]

b. good: A, [_ [ PP at NP]]
c. expert: N, [_ [ PP on NP]]
Thus in this funcion is described as the head of a PP. However,

prepositions in these positions do not exhibit syntactic properties of a
head. They take neither specifiers/attributes nor adjuncts:
(92)
a. Bill [VP believes (*right) in science (*near mathematics)].

b. Bill is [ AP good (*right) at tennis (*across the net)].
c. Bill is [ NP an expert (*right) on instruments (*close to the news)].
N o XP other than NP may follow, indicating that there is no genuine

C-selection. Neither semantically nor syntactically is a prepositional proform determined, further evidence for the absence of a referential role:
(93)
a. Bill believes in science.

*there.
*then.
therefore.
*thus.
134
GisaRauh
b. Bill is good at tennis.

*there.
"then.
*therefore.
*thus.
c. Bill is an expert on instruments.
* there.
*then.
""therefore.
*thus.
In addition, coordination with other PPs yields ungrammatical results:
(94)
a. *Bill believes in science and during his life.

b. *Bill is good at tennis and in London.
c. *Bill is an expert on instruments and in London.
Case properties, however, are present:

(95)
a. Bill believes in Mary.

her.
b. Bill is good at tennis and cricket.
them.
them.
Semantic properties of heads are absent in the sense that does not
define the type of internal argument. This argument is defined semantically by the governing V, A or :
(96)
a. Bill appealed to Mary.

*the station.
b. Bill is good at tennis.
*the station.
*the station.
However, as with the examples discussed in Section 3., theta properties

as well as properties defining the type of relation can be identified,
provided a metaphorical interpretation is assumed. Seen in this way the
prepositions from, to and in in (97) assigning the role THEME to their
external argument and the roles SOURCE, GOAL and LOCATION to
the following NP, the domains concerned here not representing spatial
135
units but rather units selected by the semantics of the governing V, A or

N:
(97)
a. The soldier died from his wounds.

b. The answers were acceptable to the committee.
c. Bill has confidence in his abilities.
Seen in this way, the prepositions on, in, and over in (98 b) to (100 b)
define the properties of the relations as do their lexical equivalents in
(98a) to (100a), i.e., SUPPORT (y, x), INCLUDE (y, x) and ABOVE
(y, x), respectively:
(98)
a. Bill sat on a chair,

b. Bill lives on rice.
(99)
a. Bill lives in a large house.

b. Bill participated in the meeting.
(100) a. The cloth lay over the table.

b. This view predominated over all the others.
The properties of in these constructions thus correspond to those
identified for in the constructions in Section 3. However, there is a
significant difference with respect to NP since as has been pointed out
there is a selectional relation between NP and the governing V, A or
N, which defines NP as the latter's object. This relation is especially
evident when passivization is possible and the NP is the subject of the
passive construction:
(101) a. Bill appealed to Mary,
b. Mary was appealed to.
(102) a. Bill talked about John,
b. John was talked about.
Examples like these illustrate that independently of there is a relation
between in this case V and NP, defined via theta role assignment
and selection properties of V, which is modified only by P. This is evident
in cases where there is an alternative between NP- and [P + NP]-complements:
(103) a. Bill believed his friend.
b. Bill believed in his friend.
(104) a. Bill knew the problem.
b. Bill knew about the problem.
136
Gisa Rauh
It may be concluded, therefore, that in these constructions is the head.

From a semantic perspective, the role of is comparable to that of case
affixes. differs from case affixes, however, as far as its position and
Case assigning properties are concerned. If, then, represents the head
of the [ + NP] constructions considered here, what is the category of
the construction as a whole? The evidence suggests that it is an NP, its
internal structure being the following:
(105)
NP
/ \
NP
This structure represents the similarity between and case affixes in

various ways. Instead of an unmarked Case, the governing N, A or
lexically assign the semantically and syntactically marked form P. thus
corresponds to inherent Cases in case-marking languages, for example,
the accusative in the context of lehren or the genitive in the context of
gedenken in German, which are also assigned lexically. The scope of is
just like that of Case the whole NP. Another correspondence is
that as well as case affixes are preserved in the context of proforms:
(106) a. Bill read Mary's book.
b. Whose book did Bill read?
(107) a. Bill gave the book to Mary.
b. To whom did Bill give the book?
Thus, to whom represents a P-marked NP, not a PP. On the basis of this
analysis a problem pointed out by O'Grady (1985) can be solved in an
elegant way.
O'Grady observes that the constructions under consideration, which
he describes as PPs, behave like NPs rather than other PPs with respect
to control and floated quantifiers. In order to capture these facts, he
introduces a new principle TDR (Thematic Dependency Requirement)
which provides a basis for distinguishing the two types of PP. One is
described as thematically dependent, i. e., it receives a theta role from the
verb. In the other case, however, it is the NP within the PP rather than
the PP that is thematically dependent. Only in the latter case may the
NP within the PP provide an antecedent for control and floated quantifiers:
(108) a. He counted on the boysi [PRO, to leave],
b. ?The book was brought by Johnj [PRO( to read during the
party].
137
(109) a. He talked to the girlsj several times each;.

b. *She sat near the boySj several times each;.
(Compare O'Grady 1985: 170, 172)
Analyzing the sequence [P -4- NP] in (108 a) and (109 a) as NP rather
than PP, as suggested here, there is no need for a new principle. The
constructions under consideration are explained with reference to government, as are other cases of NP-control (cf. Goodluck 1978) and floated
quantifiers (cf. Jaeggli 1982). In addition, structure (105) illustrates that
the dominant NP is assigned a theta role by the verb, which is then
modified by the theta role assigned by to the dominated NP.
Analyzing the [P + NP]-constructions investigated here as NPs with
as a case preposition has further advantages. It should be noted first that
the analysis is compatible with the Case Theory of the Theory of Principles and Parameters, according to which each lexical NP has to be Case
marked. Assuming that represents inherent Case lexically assigned by
the appropriate governor, the dominant NP in structure (105) is inherently
Case marked by P, whereas the dominated NP is assigned Case by P.
One problem of Case Theory as presented by Chomsky (1986 b: 192) is
the fact that of as a Case marker in the context of adjectives and nouns
is considered an exception which needs to be stipulated, (cf. Section
2.1.2.). In the analysis suggested here, the status of an "exception" is
unnecessary. When they are case prepositions in structures like (105),
prepositions in English may in general assume the function of case affixes,
i. e., in the context of V, and A, as is demonstrated by the examples
above. In the case of and A, this is the only possible way in which
Case can be marked. This assumption also provides an explanation for
the phenomenon, so far unexplained, that adjectives as well as nouns in
Modern English have lost their Case assigning properties present in Old
English. The analysis presented here supports the claim that these properties have not vanished. Instead, the semantically less differentiated case
affixes have been replaced by the more differentiated case prepositions.
The adequacy of this analysis is confirmed by the fact that in some
contexts the semantically neutral case preposition of alternates with the
genitive affix, both Case markers occurring on the periphery of an NP
rather than on the head of this NP: 16
(110) a. [ NP John]'s book
b. [ NP the man in the green shirtj's book
(111) a. the book of [ NP John]
b. the book of [ NP the man in the green shirt]
138
Gisa Rauh
Analyzing governed prepositions as case prepositions also provides an

explanation for the second or third elements in transitive phrasal prepositions. These are lexically determined by the preceding noun (e. g., with
respect to; in accordance with) or adjective (e. g., near to; next to) and in
this sense governed. As case prepositions they serve to Case mark the
following NP in a semantically differentiated way. This explains why they
do not occur in intransitive uses or with a CP-complement (cf. Section
2.3.2.).
Since Chomsky (1965: 106), and especially Hornstein and Weinberg
(1981), it has been known that the constructions considered here allow
preposition stranding, i. e., the NP may be moved out of the construction
and leave behind:
(112) a. About whom is Bill angry?
b. Who is Bill angry about?
(113) a. What did Bill comment on?
b. The book was commented on by everybody.
However, movement of NP out of a structure like (105) is excluded by
the A-over-A principle. Nevertheless, examples like (112 b) and (113) need
not be considered as evidence that (105) is inadequate. They simply
indicate that this is an unstable domain in the grammar of English,
permitting two analyses for the same sequence of units, in (112) and (113)
A + + NP and V + +NP, respectively. For V, the first of these is
presented in (105), here repeated as (114 a), the second in (114 b):
(114) a.
V'
b.
V'
NP
NP
NP
Thus, in this domain, the present state of English exhibits phenomena

which are characterized as two separate processes by diachronic linguists,
for whom a specific property of prepositions is the fact that, originating
as adverbs, they have migrated either to nouns to specify the semantics
of cases or to verbs to specify the semantics of verbs (cf., e. g., Delbrck
1893: 654 ff.; Brugmann 1911: 758 ff.; Wackernagel 1924: 153 ff.). In
addition, the examples investigated here illustrate the process described
139
by Nichols (1986: 84 ff.) as "headward migration". The form that marks

NP, a constituent dependent on V, migrates from the dependent constituent to its head. A similar process may be observed with respect to socalled phrasal verbs like look up; come o f f , carry on, etc. which, however,
will not be considered here. Examples like (112) illustrate that this analysis
has been transferred to nonverbal contexts too. It is questionable whether
a description of the syntactic ambiguity as a purely syntactic process is
justified, as Hornstein and Weinberg (1981) and others assume. The
problem seems rather to be a lexical one and is treated accordingly in
Rauh (1991 b).
5.
Summary
5.1. Prepositional variants

The preceding discussion has provided evidence for the fact that the
diverging views on prepositions presented in the introduction are justified
to some extent. None of them, however, characterizes the whole set of
prepositions, each correlates with a subset only. Thus it has been demonstrated on syntactic grounds that prepositions exhibit properties of
lexical heads and are therefore justifiably considered elements of a lexical
category by Emonds (1972; 1976); Jackendoff (1973; 1977); Stowell
(1983); Fukui and Speas (1986) or Chomsky (1986 b). The relevant properties are summarized in the following:
1. Projection properties:
lexical prepositions may take NP, CP and/or PP complements; they
may also be intransitive;
they allow pre- or postmodifying adjuncts;
they allow a specifier (if a distinction between specifier and modifier
is made).
2. Case properties:
lexical prepositions in general assign structural Case;
some deadjectival and denominal lexical prepositions, which cannot
assign structural Case, assign inherent, lexically determined Case,
expressed by "case prepositions".
3. Argument structure:
lexical prepositions are two-place predicates: (, y).
140
Gisa Rauh
4. Theta roles:
external role:
internal role:
referential role:
T H E M E (y)
LOCATION (x)
SOURCE (x)
PATH (x) or
GOAL (x)
identical to the internal role.
It was pointed out, in addition, that lexical prepositions exhibit selection al and inherent semantic properties, the former defining the properties
of potential arguments, the latter defining properties of the relation
expressed by the preposition:
5. Selectional semantic properties:
defining the ontological type of the arguments
e.g., SPATIAL (x)
T E M P O R A L (x)
defining specific properties of the internal argument
e. g., 3 DIM (x)
TIME SPAN (x)
6. Inherent semantic properties:
e.g., CONTACT (x, y)
BELOW (x, y)
I N C L U D E (x, y)
These findings provide evidence for the fact that the special treatment of
prepositions as "grammatical heads" presented by Emonds (1985), which
diverges from his earlier views, is not justified.
It has also been demonstrated that prepositions do not represent lexical
heads in all of their uses. In numerous constructions they assume other
functions. Because of these, Fillmore's statement quoted in the introduction that "preposition phrase" "is a terminological nuisance" is very
appropriate. In a subset of these constructions prepositions have precisely
the role which Fillmore identified for them in the framework of his Case
Grammar. They represent alternative, semantically specified expressions
of case affixes, "analytic cases" as Deutschbein (1917) claims, or "inflectional particles" in the terminology of Curme (1926: 29). Therefore, they
are here called "case prepositions". They are not restricted to a few, of,
to, for or by, for example, as is sometimes assumed (cf. Poutsma 1926:
761). All primary spatial prepositions and some derived ones may occur
in this function. Syntactically they form an N P together with the NP
141
which they relate to the V, A or by which they themselves are governed.

In these constructions, however, they maintain their prenominal position
as well as their Case assigning properties, and they differ from case affixes
in this respect. As semantically marked alternatives to those case affixes
which are lexically determined by a governor, they correspond to inherent
Cases, which are assigned lexically. Thus, case prepositions represent a
syntactically definable variant of prepositions and are grammaticalizations of origninally autonomous lexical forms in the sense of Meillet
(1912) and Lehmann (1985; 1986/87). Grammaticalization results in a
loss of lexical autonomy. The variants do not have lexical entries of their
own. They are introduced in the entry for the unit by which they are
governed. Their use is thus automatically restricted to complement positions. In these positions prepositions exhibit the following set of properties which, as a reduction and partial reinterpretation of the set of
properties of lexical prepositions, provides evidence for the effect of the
process of grammaticalization:
1. Projection properties
Case prepositions have no projection properties
they always precede NPs;
they allow no pre- or postmodifying adjuncts;
they allow no specifier.
2. Case properties
case prepositions assign structural Case.
3. Argument structure
case prepositions are two-place predicates: (, y).
4. Theta roles
case prepositions do not assign theta roles but the related semantic
roles, which have no syntactic relevance;
case prepositions have no referential role.
5. Selectional semantic properties
case prepositions exhibit no selectional properties defining the
ontological type of their arguments;
they do exhibit specific properties of their internal argument which
are, however, reinterpreted in accordance with the ontological type
as determined by the lexically governing category.
6. Inherent semantic properties
case prepositions exhibit inherent semantic properties which, again,
are reinterpreted in accordance with the ontological type of the
arguments as determined by the lexically governing category.
142
Gisa Rauh
Another syntactically definable variant was exemplified by those prepositions which occur in fixed phrases and form a symbiotic, syncategorematic construction with a following NP. The resulting exocentric,
marked structure X inherits syntactic and semantic properties from as
well as NP, allowing it to be used as an adjunct. Because of this combination of inherited properties, these constructions are similar to predicative adjectives and adverbs. This explains why in the history of English
predicative adjectives or adverbs have developed out of constructions of
this type. This may be taken as evidence for the tendency inherent in
natural languages to reintegrate marked structures into the grammar.
The variants considered here also represent examples of grammaticalization. Originally autonomous lexical units, they have lost their autonomy
in a symbiotic construction. Their grammatical properties in this construction are similar to those of case prepositions except that they do not
precede full NPs but fixed NP-forms and except that the reinterpretation
of selectional and inherent semantic properties occurs not in accordance
with the type of argument as determined by a lexically governing category
but as determined by the in the construction. The grammatical function
of these variants consists in attributing adjectival or adverbial properties
to NPs. The loss of autonomy means that prepositions of this type always
occur together with their nominal "Siamese sisters" in one entry in the
lexicon. In this way the markedness of these constructions is represented.
5.2. Metaphors, lexicalization and lexical units

During the course of this presentation it has been pointed out more or
less explicitly that the syntactic prepositional variants originate in metaphorical use. In conclusion, therefore, a remark on the relationship
between metaphor, lexicalization and lexical units is in place.
In connexion with lexical prepositions it was argued that lexical variants
may develop out of originally spatial prepositions via the cognitive process
of metaphor. Lexical variants represent the end of a process which starts
with a marked use, a metaphor, and is finished when the lexical properties
of the original form are reanalyzed and the result is lexicalized, when,
for example, in addition to a spatial preposition there is finally a temporal
one exhibiting its own lexical properties. The subsequent argumentation
demonstrated that this is not the only way to derive new lexical units.
Various paths may lead from metaphor to lexical unit, yielding different
results. In addition to new lexical prepositions, new lexical adjectives and
143
adverbs may be derived. This process not only involves reanalysis of

individual semantic properties but, in addition, recategorization of a
sequence of two elements of two categories to one element of one category.
In both cases forms derived from marked constructions are reintegrated
into the grammar. As new lexical elements they are autonomous and
have lexical entries of their own.
Examples of the lexicalization of marked structures were also presented,
and here there are differences in the degree of lexicalization. In Section
2.3.2. lexicalized metaphorical uses were presented as conceptual variants
of prepositions, referring to scalar and sequential uses as examples. These
are syntactically regular but semantically marked. No reanalysis but only
the reinterpretation of semantic properties is involved. The parasitic
dependency on spatial prepositions is captured by the fact that these uses
are indicated within the lexical entries for spatial prepositions, possibly
in a way suggested by Bierwisch (1983: 87 ff.) for various uses of the
same lexical unit which may be considered the result of conceptual shift.
Bierwisch rejects the notion of conceptual shift for his examples, but in
the cases discussed here conceptual shift is in fact involved; the spatial
use provides the basic meaning which represents the source of the shift.
The marked [P + NP]-constructions presented in Section 3. also represent examples of lexicalized units with lexical entries of their own. They
are, however, syntactically as well as semantically and often even
morphologically marked. Since these constructions form recurrent
patterns, their lexical representations are related by lexical rules. In
addition to such fixed phrases, there are phrasal schemata which are
semantically marked only, such as, for example, the use of spatial prepositions to express emotions as in in pain, out of hatred, etc. It may be
assumed, therefore, that there are conceptual variants here, too, which
in the process of the development of language may, but need not become
syntactically marked structures and finally develop into elements of a
different lexical category, namely adjective or adverb.
Taking into account the fact that the different kinds of lexicalization
indicated here as well as their different grammatical consequences are
due to the same metaphorical processes, it is evident that the linguistic
facts cannot be described simply on the basis of conceptual cognitive
principles which define the potential for and restrictions on metaphor.
They explain no more but also no less than the conceptual cognitive
prerequisites for the potential of lexicalization. The linguistic facts themselves, however, can only be described on the basis of grammar.
144
Gisa Rauh
Notes
1. The inadequacy of such claims is demonstrated in Rauh (1990).
2. The structural representations of the PPs are as follows:
a.
PP
I
P'
PP
I
P'
I
P
I
down
b.
his friends
PP
I
'
r
PP
I
down
I
P'
I
to
his friends
Thus in (a) to is the head of the construction and down its premodifying adjunct
whereas in (b) down is the head and to his friends its postmodifying adjunct.
3. A detailed description of the projection properties of English prepositions is presented
in Othmar (1990).
4. It should be pointed out, however, that there is a causal variant of since which takes
CP-complements only as, for example, in: He wouldn't come since he knew you were
here.
5. If one assumes, following Fukui and Speas (1986); Abney (1987); Olsen (1990) and
others, that the determiner is not the specifier of but rather the head of a DP, the
semantic relations identified here represent head-to-head relations. This assumption
does not alter the fact that these relations in this case S-selection of D with respect
to exist and are relevant. It does, however, alter the domain of the semantic
"projection properties" identified here, which are then restricted to X 1 . Inclusion of the
DP-analysis would have involved a lengthy discussion of the consequences for other
lexical categories. Given the present state of research, there would have been more
confusion than clarification, and this would have been unnecessary since the main
arguments presented here are not affected.
145
6. It should be noted here too, that applying the DP-analysis would yield different results.
According to Olsen (1990: 40), pronouns are intransitive determiners.
7. In contrast to Williams (1981 a), who assumes the theta grid implies the argument
structure, which thus need not be stated explicitly, both are here listed separately. Only
in this way is it possible to consider theta roles properties which are assigned to
arguments. And only in this way is it possible to capture the fact that there is no biunique mapping of theta roles and arguments onto each other, which is an inevitable
consequence if referential roles are assumed, as will be seen in the following.
8. Bierwisch (1988) comes to a different conclusion. He assumes that PPs are not referential
due to the fact that they do not take specifiers. In the following it will be demonstrated,
however, that with respect to P, the referential role is not related to the specifier position
as Bierwisch assumes by analogy with the referential role of N.
9. See, for example, the following, as representative of many: Bennett (1975); Brugman
(1988); Hawkins (1984); Wunderlich (1981; 1985); Herskovits (1986); Hottenroth (1991);
Lang (1991); Schepping (1991).
10. See, for example, Closs-Traugott (1985).
11. This is demonstrated by Clark (1973), for example.
12. For further information on a systematic semantic analysis of prepositions see, for
example, Wunderlich (1981; 1985) or Lang (1991).
13. See also Closs-Traugott (1985).
14. See, for example, Brugman (1988); Lindner (1981); Lakoff (1987); Langacker (1987)
or Dirven (this volume).
15. On the inheritance of properties inside compounds and derivatives see, for example,
Williams (1981 b), Lieber (1980, chapter 2; 1983) and Boase-Beier and Toman (1986).
It should be pointed out though that in the examples discussed, it is not the head that
provides the features, as is assumed by Williams.
16. This is important because English differs in this respect from case-marking languages
like German or Latin. While in these languages elements within a maximal projection
of inherit case from the case-marked head, this is not the case in English, where a
Case marker is attached to the periphery of the NP, thus exhibiting something like
phrasal inflection.
References
Abney, Stephen Paul
1987
The English noun phrase in its sentential aspect. [Ph. D. MIT.]
Akmajian, Adrian Susan Steele Thomas Wasow
1979
"The category AUX in universal grammar", Linguistic Inquiry 10: 1-64.
Baker, Marc C.
1988
Incorporation. A theory of grammatical function changing. Chicago: The University of Chicago Press.
Behaghel, Otto
1924
Deutsche Syntax. 2. Heidelberg: Winter.
Bennett, David
1975
Spatial and temporal uses of English prepositions. London: Longman Group
Ltd.
146
Gisa Rauh
Bierwisch, Manfred
1983
"Semantische und konzeptuelle Reprsentation lexikalischer Einheiten", in:
Ruzicka, Rudolf Wolfgang Mtsch (eds.), Untersuchungen zur Semantik.
(Studia Grammatica XXII) Berlin: Akademie Verlag, 15-59.
1988
"On the grammar of local prepositions", in: Bierwisch, Manfred Wolfgang
Mtsch Ilse Zimmermann (eds.), Syntax, Semantik und Lexikon. Berlin:
Akademie-Verlag, 1-65.
Boase-Beier, Joan Jindrich Toman
1986
"On -role assignment in German compounds", Folia Linguistica 22: 319340.
Bresnan, Joan
1978
"A realistic transformational grammar", in: Halle, Morris Joan Bresnan
George A. Miller (eds.), Linguistic theory and psychological reality. Cambridge/Mass.: The MIT Press, 1-59.
Brugman, Claudia
1988
Polysemy, semantics and the structure of the lexicon. New York: Garland
Publishing, Inc.
Brugmann, Karl
1911
Vergleichende Laut-, Stammbildungs- und Flexionslehre nebst Lehre vom Gebrauch der Wortformen der indogermanischen Sprachen. Bd. 2.2, von Karl
Brugmann und Berthold Delbrck. Grundri der vergleichenden Grammatik
der indogermanischen Sprachen. Straburg: Karl J. Trbner.
Chomsky, Noam
1965
Aspects of the theory of syntax. Cambridge/Mass.: MIT Press.
1981
Lectures on government and binding. Dordrecht: Foris Publications.
1986a
Barriers. Cambridge/Mass.: MIT Press.
1986 b
Knowledge of language: Its nature, origin, and use. New York: Praeger.
Clark, Herbert
1973
"Space, time, semantics and the child", in: Moore, Timothy (ed.), Cognitive
Development and the Acquisition of Language. New York, etc.: Academic
Press, 27-63.
Closs-Traugott, Elizabeth
1985
"'Conventional' and 'dead' metaphors revisited", in: Paprotte, Wolf Rene
Dirven (eds.), The ubiquity of metaphor. Metaphor in language and thought.
Amsterdam/Phila.: John Benjamins, 17-56.
Curme, George O.
1935
A grammar of the English language. 3 Volumes. Part II: Parts of speech and
accidence. Boston, etc.: D. C. Heath & Company.
Davidson, Donald
1967
"The logical form of action sentences", in: Rescher, Nicolas (ed.), The logic
of decision and action. Pittsburgh: University of Pittsburgh Press. [Reprinted
in: Actions and events. [1980] Oxford: Clarendon Press, 105-122.]
Delbrck, Berthold
1893
Vergleichende Syntax der indogermanischen Sprachen. Bd. 3.1, von Karl Brugmann und Berthold Delbrck, Grundri der vergleichenden Grammatik der
indogermanischen Sprachen. Straburg: Karl J. Trbner.
Deutschbein, Max
1917
System der Neuenglischen Syntax. Cthen.
147
Dirven, Rene
this volume.
Emonds, Joseph E.
1972
"Evidence that indirect object movement is a structure-preserving rule",
Foundations of Language 8: 546-561.
1976
A transformational approach to English syntax: Root, structure-preserving and
local transformations. New York: Academic Press.
1985
A unified theory of syntactic categories. Dordrecht: Foris Publications.
Fillmore, Charles J.
1968
"The case for case", in: Bach, Emmon Robert T. Harms (eds.), Universals
in linguistic theory. London: Holt, Rinehart & Winston, 1-88.
1969
"Towards a modern theory of case", in: Reibel, David A. Sanford A.
Schane (eds.), Modern studies in English. London: Prentice Hall Inc., 361
375.
Fries, Charles C.
1952
The structure of english. New York: Harcourt, Brace & Comp.
Fukui, Naoki Margret Speas
1986
"Specifiers and projection", MIT Working papers in linguistics 8, 128-172.
Gleason, Henry A.
1955 [1961] Introduction to descriptive linguistics. New York, etc.: Holt, Rinehart &
Winston.
1963 [1965] Linguistics and English grammar. New York: Holt, Rinehart & Winston.
Goodluck, Helen
1978
Linguistic principles in children's grammar of complement subject interpretation. [Unpublished Ph. D., University of Amherst, Amherst/Mass.]
Haumann, Dagmar
1991
-Rollen englischer Lokalprpositionen: Syntaktische und semantische Aspekte. [M. A. thesis. Bergische Universitt Gesamthochschule Wuppertal.]
Hawkins, Bruce
1984
The semantics of English prepositions. [Unpublished Ph. D., University of
California, San Diego/Cal.]
Herskovits, Anette
1986
Language and spatial cognition. An interdisciplinary study of the prepositions
Higginbotham, James
1985
"On semantics", Linguistic Inquiry 16: 547-593.
Hockett, Charles F.
1958
A course in modern linguistics. New York: The MacMillan Comp.
Hornstein, Norbert Amy Weinberg
1981
"Case theory and preposition stranding", Linguistic Inquiry 12: 55-91.
Hottenroth, Priska-Monika
1991
"Prpositionen und Objektkonzepte. Ein kognitiv orientiertes, zweistufiges
Modell fr die Semantik lokaler Prpositionen", in: Rauh, Gisa (ed.), 77107.
Jackendoff, Ray S.
1973
"The base rules for prepositional phrases", in: Anderson, Stephen R. Paul
Kiparsky (eds.), A Festschrift for Morris Halle. New York: Holt, Rinehart &
Winston, 345-366.
148
Gisa Rauh
1977
X' syntax. A study of phrase structure. Cambridge/Mass.; London: MIT Press.
1983
Semantics and cognition. Cambridge/Mass.: MIT-Press.
Jaeggli, Osvaldo
1982
Topics in romance syntax. Dordrecht: Foris.
Jespersen, Otto
1924
The philosophy of grammar. London: George Allan & Unwin Ltd.
Kittay, Eva F.
1981
"Semantic fields and the structure of metaphor", Studies in Language 5: 3163.
1987
Metaphor. Its cognitive force and linguistic structure. Oxford: Clarendon Press.
Kruisinga, Etsko
1932
A handbook of present-day English. Part II. Groningen: P. Noordhoff.
Lakoff, George
1987
Women, fire and dangerous things. Chicago; London: The University of
Chicago Press.
Lang, Ewald
1991
"A two-level approach to projective prepositions", in: Rauh, Gisa (ed.), 127
167.
1987
Foundations of cognitive grammar. Vol. I: Theoretical Prerequisites. Stanford/
Cal.: Stanford University Press.
Lehmann, Christian
1985
"Grammaticalization: Synchronic variation and diachronic change", Lingua
e Stile 20: 303-318.
1986/7
"Grammaticalization and linguistic typology", General Linguistics 26/27: 3 22.
Lieber, Rochelle
1980
On the organization of the lexicon. [Unpublished Ph. D., University of New
Hampshire.]
Lindner, Susan
1981
A lexico-semantic analysis of verb-particle constructions with up and out.
[Unpublished Ph. D., University of California, San Diego/Cal.]
Meillet, Antoine
1912
"L'evolution des formes grammaticales", Scientia (Rivista di scienza) 7: 6 24. [Reprinted in: Linguistique historique et linguistique generale. [1975] Paris:
Editions Champion, 130-148.]
Nichols, Johanna
1986
"Head-marking and dependent-marking grammar", Language 62: 56-119.
O'Grady, William
1985
"Thematic dependency and semantic interpretation", The Canadian Journal
of Linguistics 30: 159-177.
Olsen, Susan
1990
"AGR(eement) in the German noun phrase", in: Bhatt, Christian Elisabeth
Lbel Claudia Schmitt (eds.), Syntactic phrase structure phenomena in noun
phrases and sentences. Amsterdam, Philadelphia: John Benjamins.
Othmar, Martin
1990
Versuch einer systematischen syntaktischen Beschreibung der Kategorie und
ihrer Projektionen im Englischen im Rahmen eines X-bar-Modells. [. A.
thesis. Bergische Universitt Gesamthochschule Wuppertal.]
149
Paul, Hermann
1920
Deutsche Grammatik. Bd. 4. Halle: Max Niemeyer.
Postal, Paul M.
1971
Crossover phenomena. New York: Holt, Rinehart & Winston.
Poutsma, Hendrik
1926
A grammar of late modern English. Part II: The parts of speech, Section II:
The verb and the particles. Groningen: P. Noordhoff.
Quirk, Randolph
1962 [1968] The use of English. London: Longman Group Ltd.
Quirk, Randolph John Mulholland
1964
"Complex prepositions and related sequences", English Studies Presented to
R. W. Zandvoort on the Occasion of the 70th Birthday, Supplement to English
Studies 44: 64-73.
Radden, Gnter
1982
"Die bertragenen Bedeutungen der englischen Raumprpositionen", in: Radden, Gnter Rene Dirven (eds.). Kasusgrammatik und Fremdsprachendidaktik. Trier: Wiss. Vlg. Trier GmbH, (anglistik & englischunterricht 14),
133-179.
1989
"Figurative use of prepositions", in: Dirven, Rene (ed.), A user's grammar of
English: Word; sentence, test, interaction. Frankfurt am Main: Peter Lang
Verlag, 551-576.
Radford, Andrew
1988
Transformational grammar. A first course. Cambridge: Cambridge University
Press.
Rauh, Gisa
1988
Tiefenkasus, thematische Relationen und Thetarollen. Die Entwicklung einer
Theorie von semantischen Relationen. Tbingen: Gunter Narr Verlag.
1990
"Prpositionen: eine geschlossene Klasse?" Die Neueren Sprachen 89: 476498.
1991
"Prepositional forms in the lexicon: Problems and suggestions", in: Rauh,
Gisa (ed.), 169-223.
Rauh, Gisa (ed.)
1991
Approaches to prepositions. Tbingen: Gunter Narr Verlag.
Riemsdijk, Henk C. van
1978
A case study in syntactic markedness: The binding nature of prepositional
phrases. Lisse: The Peter De Ridder Press.
Riemskijk, Henk C. van Edwin Williams
1986
Introduction to the theory of grammar. Cambridge/Mass.; London: MIT Press.
Schepping, Marie-Theres
1991
"The lexical meaning of the French preposition contre", in: Rauh, Gisa (ed.),
225-252.
Stageberg, Norman C.
1965 [1977] An introductory English grammar. With a chapter on transformational grammar. New York: Holt, Rinehart & Winston.
Stowell, Tim
1983
"Subjects across categories", The Linguistic Review 2: 285-312.
Sweet, Henry
1892
A new English grammar. Logical and historical. Part I. Oxford: Clarendon
Press.
150
Gisa Rauh
Wackernagel, Jakob
1924
Vorlesungen ber Syntax. Bd. 2. Basel: In Kommissionsverlag von Emil Birkhuser & Cie.
Weinreich, Uriel
1972
Explorations in semantic theory. The Hague; Paris: Mouton.
Whitney, William D.
1889 [1969] Sanskrit Grammar. Delhi: Motilal Banarsidass.
Williams, Edwin
1981 a
"Argument structure and morphology", The Linguistic Review 1: 81-114.
1981 b
"On the notions 'lexically related' and 'head of a word'", Linguistic Inquiry
12: 245-274.
Wunderlich, Dieter
1981
"Sprache und Raum", Studium Linguistik 12: 1-19; 13: 37-59.
1984
"Zur Syntax der Prpositionalphrasen im Deutschen", Zeitschrift fiir Sprachwissenschaft 3: 65-99.
1985
"Raumkonzepte Zur Semantik der lokalen Prpositionen", in: Ballmer,
Thomas Roland Posner (eds.). Nach-Chomskysche Linguistik. Berlin, New
York: de Gruyter. 340-361.
Prepositions: Patterns of polysemization and

strategies of disambiguation
John R. Taylor
1. Introduction
Much has changed since 1973, when Ray Jackendoff could write that
"people seem never to have taken prepositions seriously" (Jackendoff
1973: 345). Not only have prepositions played a significant role in Jackendoff s own subsequent work (Jackendoff 1977, 1983), the last two
decades have seen an extraordinary burgeoning of research on this formerly neglected part of speech, by linguists working within a number of
different theoretical frameworks. 1
Several factors have contributed to the ennoblement of the humble
preposition. Foremost amongst these has been an interest in the relation
between cognition and linguistic structure, with spatial cognition, especially, being at the centre of attention. Interest in spatial cognition
coincided with the demise of what Lakoff (1987) has called "objectivist
semantics". Jackendoff (1983), Langacker (1987), Fauconnier (1985), and
many others, have argued that linguistic expressions do not and cannot
stand in any simple, direct relation to states of affairs in real or possible
worlds; rather, linguistic expressions invoke a speaker's construal, conceptualisation, or mental representation of states of affairs. Lakoff emphasises that a rejection of objectivism need not lead into the cul-de-sac
of solipsism. Conceptualisations are ultimately grounded in, and structured by, "pre-propositional" bodily experience. Our bodily sensations,
our experience of space, of objects in space, of forces acting on these
objects, provide the basic structures which, through analogy, metaphor,
and metonymy, enable us to conceptualise ever more abstract cognitive
domains. Given such a theoretical orientation, it was inevitable, perhaps,
that attention should come to focus on those lexical items whose function
is, primarily, to symbolise conceptualisations of spatial relations, and
which, through semantic extension, are extensively employed in the construal of relations in more abstract domains.
A further factor has been a renewed interest in lexical semantics, which
has led, amongst other things, to the recognition of polysemy as the
152
John R. Taylor
normal order of things in a natural language (Langacker 1988: 50), and

thereby created a need for linguistic theory to accommodate polysemy as
a natural and inevitable phenomenon (Johnson 1987:193). In this climate,
it is only natural, again, that attention should have been directed towards
those lexical items which are amongst the most highly polysemous, in
English and other languages, viz. the prepositions.
This paper focuses on some "general" patterns of meaning extension
exhibited by the English prepositions. Some meaning extensions are
idiosyncratic to individual prepositions. Such, for example, is the extension of on from its basic sense of location within a 2-dimensional plane
(There's a book on the table) to its use in connection with conveyance by
a scheduled means of transportation (There were some children on the
bus). In contrast, other processes of polysemisation are manifested, not
only by individual prepositions, but by classes of prepositions. A survey
of some general polysemisation processes is the topic of Section 3. The
survey is preceded by a brief discussion of some basic concepts in prepositional semantics (Section 2.). Then, in Section 4., I take up the issue of
disambiguation. The question is important in so far as the polysemy of
prepositions is a potential source of sentence ambiguity. Section 4. surveys
some of the strategies available to a listener for the disambiguation of
potentially ambiguous prepositional phrases.
In order to avoid possible misunderstandings, a few preliminary words
are necessary on the approach I shall take in this paper. Some readers
may balk at the idea that a preposition may have up to a dozen separate
senses; such rampant polysemy, it might be argued, ought to be contained
by a search for maximally general meanings, of which specific meanings
are contextual variants (Herweg 1989: 106; Wunderlich 1991: 593). As
Geeraerts (1992) has argued, the two approaches i.e., focus on maximally general meanings vs. focus on highly differentiated specific meanings are complementary, rather than in conflict; and indeed, Langacker's "network model" of category structure explicitly incorporates both
aspects. Thus, Langacker (1988: 51-53) envisages two kinds of relation
between the senses of a linguistic form, viz. relations of instantiation, or
elaboration (the one meaning elaborates, with greater specificity, a more
abstract, schematic meaning), and relations of extension (certain specifications of the one meaning are suspended or modified in the extended
meaning). Recognition of an array of distinct senses of a linguistic form
does not therefore in itself preclude the possibility that the different senses
might be compatible with a single more abstract representation. (But at
the same time, recognition of a single general meaning does not remove
Prepositions: Polysemization and disambiguation
153
the need to identify more specific meanings; see Taylor, 1992.) My

motivation for focusing on the more specific senses in this paper, is,
precisely, to study the general patterns of semantic differentiation exhibited by the prepositions, as well as to draw attention, when appropriate,
to departures from these general patterns.
2. Some basic notions

Prepositions denote a relation involving two or more participant entities.
The relation is inherently asymmetrical, in that one participant is selected
for foregrounding, while the other participant(s) serves as a background,
or reference point entity. The expressions in (1):
(1)
a. the picture above the sofa

b. the sofa below the picture
could well have identical truth-conditions. The expressions differ, however, with respect to foregrounding. In the former, the picture is located
relative to the sofa, in the latter, the sofa is located relative to the picutre.
Following Langacker (1987), I will refer to the foregrounded entity as
the trajector (tr) of the relation, and the reference point entity as the
landmark (lm).
In their spatial senses our main concern in this paper prepositions
characterise the spatial disposition of a tr with respect to a lm. The tr
may be either a thing, as in (2 a), or a process, 2 as in (2 b). Typically, the
lm is a thing, e.g., table in (2a) and (2b). Alternatively, the lm may be
a place, as in (2 c), where the lm of from is a place, designated by the
prepositional phrase under the table:
(2)
a. The book is on the table.

b. They were sitting on the table.
c. Take the book from under the table.
We may make a broad distinction between three categories of relations

expressed by prepositions, viz. place, goal, and path. (A fourth relation,
that of source, perhaps also needs to be added to this list; see Section
3.3.) These relations constitute what Johnson (1987) has called "image
schemas". Johnson characterises image schemas as recurring patterns of
experience involving "human bodily movement, manipulation of objects,
and perceptual interactions" (1987: xxxvi). Johnson emphasises the non-
154
John R.Taylor
propositional nature of image schemas, and at the same time their

abstractness relative to "rich" visual and kinetic images. It is, he maintains, the very abstractness and generality of image schemas that make
it possible for us to structure "indefinitely many perceptions, images, and
events" (p. 29). Consider the notion of "place". That things occupy places
relative to other things is a basic, recurrent element of our experience.
Yet the notion of place is sufficiently abstract as to preclude specification
of the precise spatial disposition of one thing with respect to another.
Objects not only occupy places, they also move from one place to
another place, along a trajectory. Leaving aside the possibility of perpetual
motion, any object moving on a trajectory has necessarily started at some
place of origin, and will come to rest at a place of destination. Likewise,
arrival at a destination necessarily presupposes movement along a trajectory, from some place of origin. Thus the notions of origin, trajectory
and destination mutually presuppose one another. However, different
linguistic forms can focus on, or "profile", different facets of the objecton-a-trajectory schema. In a path sense, a preposition profiles only the
trajectory (or a part of the trajectory) of a tr, without regard for the endpoint or point of origin; conversely, in its goal sense, a preposition profiles
only the end-point of a trajectory, both the origin and the trajectory itself
being unspecified and out of focus.
The notions of place, goal, and path are thus intimately linked, in
virtue of the trajectory schema; a goal, for example, is a place at the
termination of a path. It is not surprising, therefore, that single lexical
items, viz. the prepositions, exhibit considerable polysemy with regard to
the three notions. We shall see, however, that further refinements on this
three-way distinction will prove to be necessary.
3. Polysemisation processes
The following prepositions were take into consideration for the purpose
of this study:
above, across, along, (a)round,3 at, behind, beside, between, beyond, by,
down, from, in, in front of, into, o f f , on, onto, out (of), over, past, through,
throughout, to, towards, under, up, via.
155
3.1. PLACE
With the exception of via, to and its compounds (towards, into, onto),
and past (whose idiosyncratic properties will be mentioned in due course),
all the prepositions under consideration can designate the place of a tr
with respect to a lm. Place is construed as a stative relation, which can
persist without change for an unbounded period of time.
With respect to their place meanings, the prepositions can be divided
into two broad classes.
(i) Simplex. The majority of the prepositions belong to this class. The
place of the tr is construed as an internally unstructured point, whose
location is characterised relative to the lm. The different prepositions
locate the tr in different ways relative to the lm. In locates its tr at a
place enclosed by a bounded area or volume, out of locates its tr at a
point external to a bounded area or volume, and so on.
Simplex prepositions may take as their tr not only entities which are
"punctual" in a strict geometrical sense, they also permit spatially extended trs (the book on the table) and multiplex trs (the tr, that is, consists
of a multiplicity of entities, e. g., the books under the table).4 The tr may
also be a process. The process may likewise involve a simplex participant
(He was sitting at the table) or a multiplex participant (They were sitting
at the table).
(ii) Multiplex: Multiplex prepositions include (a)round, along, through,
throughout, as well as, in some of their senses, up, down, over and between.
The place at which the tr is located is construed as a multiplicity of points
which are occupied simultaneously by the tr. There are two possibilities.
Either the tr is a spatially extended entity, whose component parts
simultaneously occupy the points; or the tr is a multiplex entity, whose
component members simultaneously occupy the points:
(3)
a. There's a fence round the garden, (extended tr)

b. There are trees round the garden, (multiplex tr)
Alternatively, the tr may be a process that is instantiated simultaneously

at a multiplicity of points located relative to the lm:
(4)
People were talking round the table.
Multiplex prepositions impose on their tr a conceptualisation consistent

with the notion of a multipunctual place. While the following sentences
could well describe the same scene, the scene is conceptualised in different
ways:
156
(5)
John R. Taylor
a. There's a row of trees by the river, (simplex place)

b. There's a row of trees along the river, (multiplex place)
In (a), the row of trees is conceptualised as a spatially extended entity

located at an internally unstructured point. In (b) the row of trees is
located at a multiplex place, which is occupied simultaneously by each
constituent member of the tr. Semantic oddity results if the tr cannot be
conceptualised in a manner consistent with the requirements of the
preposition:
(6)
* There's a tree along the river.

* There's a man sitting round the table.5
A number of multiplex prepositions impose a further requirement on

the tr. Compare:
(7)
a. There's a row of trees along the river.

b. There's a road along the river.
(8)
a. There's a row of trees up the hill.

b. There's a road up the hill.
In both sentences in (7), the tr of along occupies a multiplicity of points

located relative to the length of the river. In both sentences the tr occupies
a multiplex place. In (8 b) the tr likewise occupies a multiplex place; the
road, that is, extends from the bottom of the hill, up the side of the hill,
to its top. But in (8 a), the tr occupies a simplex place, located at the end
of an imaginary path (see 3.5); the trees, that is, are located on the top
of the hill. The reason for the difference between (8 a) and (8 b) lies with
the nature of the tr. A road, like a row of trees, may be conceptualised
as an extended, one-dimensional (1-D) entity. Alternatively, a road typically defines a potential path, which may be followed by an object
moving in a certain direction. Other 1-D entities which define a potential
path include footpaths, bridges, tunnels, and railway lines. Significantly,
(8 a) can acquire a reading analogous to (8 b) through the addition of
elements which convey the idea of directed motion along a path:
(9)
a. There's a row of trees going up the hill.

b. Follow that row of trees up the hill.
Like up, the prepositions down, through, over, and perhaps more marginally, past, can denote the multiplex place of an extended 1-D tr only
on condition that the tr can define a potential path. (The asterisked
157
sentences in (10) are, of course, acceptable on an end-point of a path

reading, to be discussed in Section 3.5.)
(10)
a.
b.
c.
d.
There's
There's
There's
There's
(a tunnel/*a vein of ore) through the mountain.

(a way I* a row of houses) down the hill.
(a footpath)*a fence) over the hill.
(?a main road/*a row of trees) past our house.
Again, multiplex readings of the asterisked sentences become possible

if some element in the sentence suggests the notion of directed motion:
(11)
a.
b.
c.
d.
There's
There's
There's
There's
a vein of ore running through the mountain.

a row of houses going down the hill.
a fence going over the hill.
a main road running past our house.
Between and over profile both simplex and multiplex relations. In (12),
the tr occupies a simplex point relative to the lm entity(ies):
(12)
a. There's a tree between the two houses

b. There's lamp hanging over the table
(13)
a. There's a fence between the two houses

b. He wore a pullover over his shirt
In (13) the tr simultaneously occupies a multitude of points located

relative to the lm. 6 Note that the two uses of over in the (b) sentences
correlate with other, idiosyncratic differences. In (12 b), the tr must be
vertical to, but not in contact with the lm. In (13 b), the tr is in contact
with the lm, and need not be vertical to it. 7
3.2. Multiplex prepositions: PLACE and PATH

As a group, multiplex prepositions permit a range of semantic extensions
which are not available for simplex prepositions.
(i) As already noted see (3) the place denoted by a multiplex
preposition may be occupied simultaneously, either by the parts of a
single, extended tr, or by the components of a multiplex tr. This relation
may be expressed informally as follows:
(A) (with multiplex prepositions) Extended Tr <- Multiplex Tr 8
(ii) The points invoked by a multiplex preposition may be occupied
simultaneously (by an extended or multiplex tr), or successively, i. e., by
158
John R. Taylor
a moving entity. In this latter case, the tr traces a path whose trajectory
is characterised relative to the lm.
(B) (with multiplex prepositions)
Place of Tr -> Path of Tr
Verbs of motion naturally impose a path reading on a following preposition:
(14)
a. There are trees along the river, (place)

b. We walked along the river, (path)
(15)
a. She wore a necklace round her neck.

b. The Earth moves round the Sun.
See which Jackendoff treats as a verb of motion also imposes a

path reading, as does look:9
(16)
a. There's a bridge across the harbour, (place)

b. We can't see across the harbour, (path)
(17)
a. There's a road going up the hill.

b. Look up the hill.
A path may be perfective or imperfective. A perfective path terminates

upon the tr's arrival at a definite end-point. Perfectivity may be suggested
by the use of the adverbials right, all the way, half way, etc. If you walk
right round the block, your path terminates upon your arrival at your
original point of departure, if you drive all the way through the tunnel the
path terminates on your emergence from the tunnel, if you swim half way
over the river the path terminates at a midpoint in the river. An imperfective path, in contrast, is of indeterminate extent. Drive through the
tunnel, swim over the river, walk across the street, etc. could denote any
arbitrary portion of a path. These prepositions can also suggest an
iterative path. A jogger may run round {and round) the block an indefinite
number of times. An iterative path is a special instance of an imperfective
path.
In their path senses, over and (a)round have undergone further specialisation, in that the lm with respect to which the path is characterised
may be conceptualised as an obstacle that the tr must "get over" or "get
round" in order to pursue its journey. Thus (18) may be interpreted in
(at least) two ways. It can convey that the tr traces a roughly circularshaped path with respect to the base of the mountain; alternatively, the
159
tr makes a semi-circular deviation in its passage from some unspecified

starting point to some unspecified destination.
(18)
We drove round the mountain.
(iii) A further source of polysemy with multiplex prepositions is the

dimensional properties of the configuration of points profiled by the
preposition. The points may form a linear, one-dimensional configuration,
or they may be distributed in two- or three-dimensional space.
(C) (with multiplex prepositions)
1-D Tr <- 2-D Tr *- 3-D Tr
There are some substantial differences amongst multiplex prepositions
in this regard:
Along, across, through, up, and down: only a 1-D configuration
Over: 1- and 2-D configurations
Throughout: 2- and 3-D configurations
(A)round and between: 1-, 2-, and 3-D configurations 10
Extension process (C) interacts with (A) and (B). (19) and (20) illustrate
some of the possibilities with regard to over and throughout.
(19)
a. There's a footpath over the hill, (place; extended 1-D tr defining

a potential path)
b. He walked over the hill, (path linking 1-D configuration of
points)
c. There are trees dotted all over the hill, (place; multiplex tr; 2D)
d. There's water all over the floor, (place; extended tr; 2-D)
e. He walked all over the city, (convoluted path, linking 2-D
configuration of points)
(Note that when it profiles a 2-D configuration of points, over must be

preceded by all.)
(20)
a. Soldiers were posted throughout the city, (place; multiplex tr; 2D)
b. There's wall-to-wall carpeting throughout the house, (place; extended tr; 2-D)
c. There was smoke throughout the building, (place; extended tr; 3D)
d. We searched for him throughout the building, (path, linking up
a 2- or 3-D configuration of points)
160
John R. Taylor
Especially complex is (a)round. In addition to the various dimensional

tr configurations permitted by (a)round, there is also the possibility that
the tr may be located either external to, or internal to the boundary of
the lm. In the case of a 1-D tr, the location of the tr may also coincide
with the lm boundary. (21) therefore has (at least) three interpretations.
The trees may be planted along the boundary of the garden; they may
be planted external to the garden; or they may be planted inside the
garden.
(21)
There were trees planted round the garden.
With a 1-D tr, a further source of complexity concerns the distinction

between a perfective configuration of points (i. e. a configuration which
"joins up"), and an imperfective (either partial or iterative) configuration.
Compare:
(22)
a. I must have put on weight. This belt won't even go round my

waist.
b. He put his arm round her waist.
c. Wind the string round your finger.
(22 a) means that the tr is not long enough to completely encompass the
circumference of the lm (perfective path); (22 b), however, is appropriate
even though the man's arm does not completely encompass the girl's
waist (partial, imperfective path); while (22 c), depending on the length
of the string, could convey an indefinite number of revolutions (iterative,
imperfective path).
With over and (a)round, the points in a 2-D configuration are often
construed as being randomly distributed within a bounded region. 11
(23)
a. The guests were standing round the room, talking.

b. He went round the guests, introducing himself.
In (a), the multiplex tr simultaneously occupies a set of points randomly

distributed in a 2-D area circumscribed by the boundary of the lm. In
(b), the tr traces a path which links up randomly distributed points. The
notion of random distribution also motivates certain "intransitive" 12 uses
of (a)round. In (24), the lm of the preposition is not explicitly stated;
probably, the lm would be understood to be the environment of the
speaker. The tr has been located at randomly selected points relative to
this unspecified lm.
(24)
a. This product's been around for many years.

b. He's been hanging around for too long.
161
Towards does not participate in the semantic extensions discussed in

this section. Towards profiles only an imperfective path. It invokes a 1D configuration of points relative to its lm, but these points can only be
occupied successively, not simultaneously.
(25)
a. We drove towards the city.

b. * There were trees towards the city.
c. *?There is a road towards the city.13
3.3. PLACE and GOAL

In general, prepositions that denote the place of a tr can also denote a
goal, i. e., a place which the tr comes to occupy with respect to the lm.
The goal reading is permitted irrespective of whether place is construed
as a simplex point, or as a 1-, 2-, or 3-D configuration of points.
(D) Place of Tr -> Goal of Tr
A couple of examples:
(26)
a. The lamp is above the table, (place)

b. Hang the lamp above the table, (goal)
(27)
a. There's a fence around the garden.

b. They erected a fence around the garden.
Not unexpectedly, place-goal polysemy can give rise to ambiguities. In

(28 a) the jumping could be located at a place "on the wall"; alternatively,
a place "on the wall" could be the goal of the jumping.
(28)
a. He jumped on the wall, (place or goal)

b. She fell in the water.
Of somewhat unclear status with regard to place-goal polysemy are

off out (of) and from. On the one hand, these have been classified as
prepositions of source (e.g., Dirven 1989), on the grounds that in sentences like those in (29), the lm entity characterises a place that the tr
moves away from:
(29)
a. We drove off the main road.

b. Mandela came out of prison.
c. John has just arrived from America.
162
John R. Taylor
But it can also be argued that these sentences indicate the place that the
tr comes to occupy, i. e., the prepositions can be legitimately classified as
prepositions of goal. In (a), for example, the tr comes to occupy a place
located "not on lm". Consistent with (D), o f f , out (of), and from can also
be used merely to indicate place, i. e., a place not on lm, a place not in
lm, or a place separated from lm.14
(30)
a. The motel is off the main road.

b. Mandela is out of prison.
c. John is from America.
Of the prepositions under consideration, to and its compounds (into,

onto, and towards) fail to exhibit place-goal polysemy. As already noted,
towards is an exclusively path preposition, which can denote neither place
nor goal. To serves exclusively as a goal preposition, and conveys that a
tr comes to be located at a lm.
(31)
a. The children have gone to school.

b. The children are now (at/*to) school.
Likewise into and onto convey that the tr comes to be in or on the lm.
Given the place-goal polysemy of in and on, into and onto in (32 a) and
(33 a) may be replaced by the monomorphemic prepositions in and on.
(32)
a. I've put the money (into/in) my pocket.

b. The money is now (inl*into) my pocket.
(33)
a. I've put the vase (onto/on) the ground.

b. The vase is now (on/*onto) the ground.
In (32 a) the use of into rather than in emphasises the fact that the tr
crosses the boundary of the container lm. But into can also convey that
a tr penetrates more deeply into the lm, without necessarily first crossing
a boundary; in this case, into is not replaceable by in.
(34)
We drove deeper (into/*in) the forest.
While at is predominantly a place preposition, it has limited and

specialised use as a preposition of goal. In this sense, at means, not that
a tr comes to be located "at" a lm, but that a tr is aimed at, and/or
forcefully propelled towards a lm.
(35)
They threw stones at the police.
163
3.4. Simplex prepositions: PLACE and PATH

We have seen that multiplex prepositions can designate both a place and
a path (3.2.). A small number of simplex prepositions viz. by, behind,
in front of, under and between can likewise designate both a stative
place relation and a path. In the latter case, the prepositions identify a
place which a tr passes through or passes by on its trajectory from some
unspecified place to another unspecified place.
(E) Place of Tr <-> Place on Path of Tr
The following sentences are therefore three-ways ambiguous; in addition
to the place and goal readings cf. (25) the prepositional phrases
can also denote the path of the tr:
(36)
a. The dog crept under the table.

b. The mouse ran behind the curtain.
c. The child walked in front of me.
Via is reserved exclusively to denote a place construed as a point on a

path. Place as a point on a path is also the predominant meaning of past.
Over and through can likewise designate a place on a path:
(37)
a.
b.
c.
d.
I went to Cape Town via Bloemfontein.

We must have driven past his house.
Throw the ball over the wall.
He drove through a red light.
The following pairs of prepositions contrast with regard to their ability

to designate a place as a point on a path:
PLACE and PATH
by
in front of
behind
between
over
PLACE only
next to, beside
at the front of
at the back of
in between
above
Compare, for example, the following pairs of sentences (the asterisks

denote unacceptability on the intended senses):
(38)
a. We drove (byj*next to) the post office, (path)

b. We live (by/next to) the post office, (place)
164
John R. Taylor
(39)
a. The car drove right (in front off* at the front o f ) me. (path)
b. They were playing (in front of/at the front o f ) the house, (place)
(40)
a. They ran (between)*in between) the two trees, (path)

b. They were playing (between/in between) the two houses, (place)
(41)
a. He jumped (over/*above) the wall, (path)

b. The lamp hangs (over/above) the table, (place)
In and out have specialised uses as path prepositions, exemplified in

(42).
(42)
a. The burglar came in the back door.

b. The cat jumped out the window.
Here, the lm entities, i. e., the back door and the window, are construed
as apertures in the perimeter of an enclosure, through which the tr enters
or exits (cf. Hawkins 1986). The use of in and out exemplified in (42) is
idiosyncratic to English, and some other Germanic languages (Taylor
1991). While one can say, in German, Die Katze ist aus dem Fenster
gesprungen, in French the path function of the lm entity would have to
be explicitly encoded by means of the path preposition par.
(43)
Le chat a saute (par/*hors de) la fenetre.
In English, also, the path function of the lm may be made more explicit:
(44)
a. The burglar came in by the back door.

b. The cat jumped out through the window.
Here, by the back door and through the window designate a place on the
path of the tr, while in and out are used intransitively, their lm being
understood to be the 3-D enclosure in whose perimeter the back door
and the window are located.
3.5. PLACE and GOAL as end-point of PATH

Prepositions which denote the path of a tr can also locate a tr at a place
construed as the end-point of a path that an observer would have to
trace in order to reach the tr.
(F) Path -> Place construed as end-point of Path
Via, by and towards do not undergo extension (F). Otherwise, as the
following examples testify, (F) is extremely productive.
165
(45)
a. John drove round the corner, (path)

b. The post office is round the comer, (place as end-point of path)
(46)
a. John swam across the river.

b. John lives across the river.
(47)
a. They walked under the bridge.

b. The railway station is under the bridge.
(48)
a. The dog ran through the kitchen.

b. The pantry is through the kitchen.
(49)
a. They ran down the road.

b. Their house is down the road.
(50)
a. I swam over the river.

b. They live over the river.
(51)
a. Drive past the police station.

b. The school is past the police station.
When denoting place as end-point of a path, the prepositions typically

have a strongly deictic component, i. e., they imply a path which has its
origin at an observer, typically the speaker or hearer. (45 b), for example,
conveys: "In order to get to the post office, you have to go round the
corner (from here)." The origin of the imaginary path may be explicitly
stated:
(52)
The post office is round the corner from the police
station.
Most prepositions that denote a place can also denote a goal (section
3.3.). Not surprisingly, therefore, the end-point of a path may be presented
either as the place of the tr, or the goal towards which the tr moves.
(G) Path of Tr - Goal of Tr at end-point of Path
If we take into account this second possibility, it will be apparent that
the (a) sentences in (45)(51) are subtly ambiguous. John swam across the
river could designate the path which the tr traces as he swims. The
sentence could focus also on the goal of the swimming, i. e., a place
located "over the river".
The expressions over here and over there exemplify an idiomatic extension of the end-point sense of the preposition. The expressions direct
attention to a place which is in the perceptual field of both speaker and
hearer, and which is construed as the end-point of an imaginary path
166
John R. Taylor
whose origin is located at the addressee. The plane of the imaginary path
must approximate to the horizontal (otherwise up here, down there, etc.
are used.) The end-point sense also underlies the intransitive use of over
and round in (46):
(53)
He came round/over to see me.
Here, the lm of the

understood to be the
the tr's goal, i. e., the
the tr traces a path
intervening space.
preposition is sub-lexical; it would probably be

space separating the initial location of the tr and
home of the speaker. In order to reach this goal,
of unspecified topology "round", or "over" the
3.6. PLACE as RESULT

Closely related to, but distinct from, the end-point focus discussed in
3.5., is the resultative use of prepositions. In the former case, the preposition profiles the end-point of a potential path, in the second case, the
preposition profiles a place relation which has resulted from the attainment of a goal. Compare:
(54)
a. He lives over the hill.

b. The horse is over the fence!
(54 a) locates the tr at a place construed as the end-point of an imaginary

path that originates with an observer. The sentence, as we have seen, is
strongly deictic; note also that in (a) the tr itself is not involved in motion
along a path. (54 b) might be uttered during a commentary on a horserace. The sentence has no deictic component, and locates the tr with
respect to a goal at the end-point of a path which the tr itself has
traversed.
(H) Goal of Tr <- Place of Tr as Result of Tr reaching Goal
Result is an extension of goal, not of path. Thus, into and onto, which
can denote a goal but not a path, can be used in a resultative sense.
(55)
a. Everything is now into position. ( = Everything has been placed

in position)
b. The actors are already onto the stage.
167
Certain non-spatial uses of into and onto need to be analysed as result.

The prepositional phrases in the following sentences denote stative relations, the result of the achievement of a goal.
(56)
a.
b.
c.
d.
I worked until late into the night.15

We're already into January.
He's into spiritualism.
I'm onto something big.
4. Disambiguation
The polysemy of prepositions as of any linguistic structure is a
potential source of sentence ambiguity. Given the reasonable assumption
that the majority of lexical items in a language are polysemous to some
degree, one might expect that an increase in sentence length would be
accompanied by an exponential increase in the number of possible interpretations of the sentence. Yet, in general, a sentence is not ambiguous
to a degree consistent with the polysemy of its constituent units. One
does not come across sentences that are, say, seven-ways ambiguous, on
account of the seven senses of a preposition, or sentences that are 32ways ambiguous on account of five constituent morphemes each with
two established senses. Clearly, in the process of sentence comprehension,
only a restricted number of senses perhaps even only one sense of
a lexical item is selected. In this final section, I want to consider some of
the factors which might guide a listener in this process. For without
procedures for selecting from among the established senses of a lexical
item, polysemy would lead to rampant ambiguity, which in turn would
severely compromise the communicative efficiency of the language code.
For this purpose, I will focus on some sentences with over. Over has a
wide range of distinct senses. It may profile both a simplex and a multiplex
relation; as a place preposition, its tr may be located with respect to a
punctual place, or to a multipunctual 1-D or 2-D place. In addition, over
may designate a goal, a perfective or imperfective path, or a place or
goal construed as the end-point of a path. How does a listener, presented
with an instance of over, narrow down the range of senses that might be
implicated in the interpretation of the sentence?
(i) A first, important factor is a person's knowledge of how the world
(normally) is. Often, certain senses of a preposition can be rejected
168
John R. Taylor
outright because of the pragmatic or factual implausibility of the resultant

sentence meaning. Compare the sentences in (57):
(57)
a. The helicopter hovered over the city.

b. The boy jumped over the wall.
In (a), over would be taken as a simplex place preposition; the hovering

of the helicopter is located at a place vertical to, but not in contact with,
the city. An analogous interpretation of (57 b) cannot be seriously entertained; it would be in defiance of gravity for the boy's jumping to be
located at a place vertical to, but not in contact with the lm. An alternative
interpretation must therefore be sought, e.g., over the wall is taken to
designate the path of the tr.
(ii) A second factor is the valence (or "thematic structure") of other
constituents in the sentence, especially the verb. 16 Compare put and throw.
Both verbs denote activities which are directed towards bringing an entity
into a certain spatial disposition. Additionally, throw conveys that the
entity is propelled on a path, a possibility which does not exist for put.
Hence the prepositional phrase in a sentence of the form [NP-/jm/-NPPP] can only be interpreted as an expression of goal, while the prepositional phrase in [ N P - f / i r o w - N P - P P ] may be interpreted as an expression
of goal or as the designation of a path, or even as an expression of goal
construed as the end-point of a path. Accordingly, (59) is open to
interpretations that are not available for (58).
(58)
John put his coat over the chair.

(i) the coat comes to cover the chair
(59)
John threw his coat over the chair.

(i) the coat comes to cover the chair
(ii) the coat moves on a trajectory, part of which is vertical to
the chair
(iii) the coat comes to occupy a place at the end of the trajectory
characterised in (ii), i. e., a place on the other side of the chair
with respect to an observer
(iii) A third factor is the aspectual character of the situation denoted by

a sentence. Aspect is important in so far as the notion of place implies a
stative relation, while goal implies a perfective process. Paths, as we have
seen, may be either perfective or imperfective. If the reaching of an endpoint is in focus, we have to do with a perfective path; otherwise, the
path is imperfective.
169
The following sentence, without further context, is ambiguous:

(60)
The plane flew over the Atlantic.
On an imperfective reading, the sentence merely conveys that the plane's

flight was located vertical to the Atlantic, while a perfective reading
entails that the plane crossed the Atlantic. The sentence may be disambiguated by the addition of the phrases for amount of time and in
amount of time; the former presupposes imperfectivity, the latter perfectivity (Vendler 1967):
(61)
a. The plane flew over the Atlantic for five hours.

b. The plane flew over the Atlantic in five hours.
Other factors may impose an imperfective reading. In (62):

(62)
The boy jumped over the wall for five
minutes.
a place interpretation of over the wall would be consistent with the

imperfectivity presupposed by for five minutes. Yet, as already noted in
connection with (57 b), a place interpretation is ruled out on grounds of
factual impossibility. One way of making sense of (62) is to invoke an
iterative reading: "The boy repeatedly jumped back and forth over the
wall." Or consider (63):
(63)
The plane flew over the Atlantic for ten years.
On pragmatic grounds, we again reject an interpretation of (63) analogous

to (61 a): it is not technically feasible for a plane to be located at a place
vertical to the Atlantic for ten years at a stretch. We make sense of (63)
again by invoking an iterative reading: "The plane repeatedly crossed the
Atlantic", i. e., "The plane was in service on a trans-Atlantic route."
(iv) The interpretations accorded to (62) and (63) testify to a complex
interplay of prepositional meaning, aspect, and factual knowledge. But a
further factor seems to be involved, namely informativity. Informativity
is to be understood in Gricean terms, i. e., a listener interprets a sentence
against the assumption that the speaker only provides an amount of
information consistent with his communicative intentions. 17
Compare the interpretations accorded to the following sentences:
(64)
a. I flew over the Atlantic for five hours.

b. I went over the street for five minutes.
In spite of their structural parallelism, these sentences are interpreted in

radically different ways. The difference has to do with the scope of the
170
John R.Taylor
temporal phrase. 1 8 In (64 a), for five hours has as its scope the VP "fly
over the Atlantic", i. e., for five hours denotes the duration of the flight.
In contrast, for five minutes in (64 b) has only the locative phrase "over
the street" as its scope, i.e., for five minutes denotes the amount of time
spent at the destination. (64 a) does NOT mean that I flew to a place on
the other side of the Atlantic, and stayed there for five hours. Likewise,
(65 b) does NOT mean that I spent five minutes engaged in the activity of
going over the street.
W h a t triggers the different scope interpretations in (64)? One possibility
is the pragmatic plausibility of the preferred interpretations, and the
corresponding implausibility of the rejected interpretations. Against this
hypothesis, note that it is difficult to "force" the rejected interpretations
by inserting pragmatically more plausible temporal phrases. Although
intuitions concerning (65) are not quite as clear as with (64), the temporal
phrases in (65) still seem to prefer VP scope and P P scope, respectively.
(65)
a. I flew over the Atlantic for six months.

( duration of stay at goal) 1 9
b. I went over the street for twenty seconds.
( duration of activity)
A second possibility might be that the different interpretations are

triggered by differences in the valence of the verbs go and fly, i.e., go
preferentially invokes a perfective path terminating at a goal, while fly
preferentially invokes an imperfective path. Thus, in association with \go
+ locative phrase], a phrase of the kind for amount of time, which
presupposes imperfectivity, is not understood to refer to the temporal
duration of the process itself. The listener must search for an alternative
interpretation, one possibility being that the temporal phrase concerns
the a m o u n t of time the tr was located at the goal.
Consistent with this hypothesis, we note that the temporal phrase in
(66) can only have P P scope:
(66)
I went to Paris for five days.

( M y journey lasted five days)
However, the hypothesis is rendered suspect by the fact that the temporal
phrase in (67), with the verb fly, also has PP scope:
(67)
I flew to Paris for five hours.

( My flight lasted five hours)
171
The possibility that these interpretations reflect certain properties of the

-phrase must also be rejected. The temporal phrases in (68) seem to
permit both VP and PP scope.
(68)
a. I travelled to Paris for five days.

b. I drove to Paris for five hours.
(five hours = duration of journey to goal OR duration of stay
at goal)
Also, certain expressions with go seem to permit both kinds of interpretation.

(69)
I went round the island for five days.

(I spent five days going round the island, OR I went to a place
"round the island" and stayed there for five days.)
On the other hand, the /or-phrase in (70) only has PP scope.

(70)
/ went round the corner for five minutes.
An alternative explanation of the highly complex state of affairs exemplified by these examples appeals to the notion of informativity. (That
intuitions concerning these sentences do seem to vacillate, could be
construed as circumstantial evidence that the explanation offered here is
on the right track, informativity being a quintessentially pragmatic, and
therefore variable notion.) Consider, first, (64). Both fly and go denote
motion on the part of the tr. But fly is by far the more informative word,
in that it also specifies, not only motion, but also the manner of the
motion. A speaker's choice of fly rather than go can thus signal his
intention to focus on the motion of the tr in its own right. Consistent
with this intention, a locative phrase, such as over the Atlantic, is most
reasonably interpreted as giving further information on the motion of
the tr, i.e. over the Atlantic designates the place or path of the tr, rather
than its goal. The temporal phrase for amount of time will likewise be
interpreted with regard to the tr's place or path. On the other hand, the
selection of a semantically less informative motion verb, such as go, is a
clue to the speaker's intention to focus, not on the motion of the tr per
se, but on some other aspect of the trajectory, e. g. the destination. Thus
a locative phrase is likely to be construed as an expression of goal, while
a temporal phrase of the kind for amount of time will be taken to refer
to some aspect of the tr's location at the goal. Travel and drive, in (61),
being of intermediate informativity in comparison with go and fly, permit
both kinds of interpretation.
172
John R. Taylor
Why, then, is it possible to interpret (69) analogously to both (64 a)

and (64 b)? The answer, I suggest, lies in the relative uninformativity, as
an expression of place, of the phrase round the island. As an expression
of place construed as the end-point of a path, round the island could
denote virtually ANY spot on the perimeter of the island. While (62) is
open to the interpretation that the tr went to some arbitrary spot and
stayed there for five days, the very uninformativity of the locative phrase
invites an alternative interpretation, i. e., round the island may denote the
path of the tr, and the temporal phrase is understood relative to the
travelling of the tr, not to the eventual destination.
Notes
1. For treatments within Cognitive Linguistics, see especially Brugman (1988) (expanded
upon in Lakoff 1987: 416-461); Hawkins (1984); and Vandeloise (1984); and shorter
treatments by Dirven (1989); Radden (1989); and Schulze (1987, 1989). For a study of
prepositions within the theoretical framework of Jackendoff, see Cienki (1989). Important contributions have also been made by Bierwisch (1988); Herweg (1989); and
Wunderlich (1991).
2. "Thing" and "process" are used here as technical terms, in the sense of Langacker
(1987). The same applies to "profile", used below.
3. Round and around are often interchangeable, although in some contexts there are clear
preferences for one or the other of the two prepositions. For the purpose of this study,
the two words were treated as synonymous.
4. Spatially extended and multiplex trs might appear incompatible with the notion of
punctuality, implicit in the notion of place. The crucial feature of the construct "point"
is, however, its lack of internal structure, rather than its strictly geometrical properties.
5. These sentences would, however, be acceptable if the place of the tr were construed as
the end-point of a path (see 3.5.).
6. Multiplex between specifies two different configurations of points: either the points link
up the lm entities, i. e., in (13 a) the fence goes from the one house to the other, or they
may separate the lm entities, i. e., the fence may stand on the boundary of the plots of
land on which the two houses are built.
7. In some of its uses, around can profile the place of a simplex tr {He lives somewhere
around here). Some complex prepositional phrases (e. g .from X to Y) profile a multiplex
place: There's a motorway from London to Oxford.
8. In order not to commit myself to the "directionality" of semantic extention this
would involve taking a stand on the question of which of the two meanings is more
"basic" than the other semantic extensions are represented by a double arrow <->.
9. Jackendoff (1985: 150) gives as a rough paraphrase of sees y "x's gaze goes to y."
Likewise, looks at y may be paraphrased "x sends his gaze to y."
10. As with the notion of "point", so also dimensionality is a matter of conceptualisation,
rather than geometry. Strictly speaking, a wooden board is a three-dimensional entity.
When you hammer a nail into a board, the three-dimensionality of the board is in
11.
12.
13.
14.
15.
16.
17.
18.
19.
173
focus. But when you lay a board across, or along a footpath, the board is conceptualised
as a linear, one-dimensional entity. If the board is in the garage, the board is conceptualised as a zero-dimensional, i. e., punctual entity.
Some speakers appear to distinguish round and around mainly with respect to this
aspect, around being preferred for a randomly distributed location.
The use is "intransitive" in that the preposition is not followed by an "object", i.e., a
nominal designating a lm. Traditionally, around in (24) would be categorised as an
adverb, or particle, not a preposition. Jackendoff (1973) advances a number of arguments for the categorial unity of the traditional class of prepositions, certain adverbials,
and certain so-called particles (as well as certain conjunctions). His arguments had
been prefigured by Jespersen (1924: 87 ff.). In cognitive grammar terms, we would say
that the lm of the intransitive prepositions in (24) is sub-lexical; as noted in the text,
the lm would probably be understood to be the environment of the speaker.
(25 c) is perhaps marginally more acceptable than the (a) and (b) sentences, since the
notion of "road" suggests a potential path.
Note that from need not imply any movement on the part of the tr away from the lm.
While travelling towards a destination, we may say that we are still 100 km from the
destination.
The affinity of result and goal is suggested by the German equivalent of (56 a): Ich
arbeitete bis spt in die Nacht. Note that in takes accusative case (the case associated
with the goal sense of the preposition), not dative case (the case associated with the
place sense).
On "valence", see especially Langacker (1987: 277 ff.).
Cf. Grice's maxims of quantity (Grice 1975):
1. Make your contribution as informative as is required;
2. Do not make your contribution more informative than is required.
I am indebted to Dieter Wunderlich (p.c.) for this formulation.
The most plausible interpretation of this sentence is based on an iterative reading of
the VP, i. e., "For six months I repreatedly flew over the Atlantic." Note that I flew
overseas for six months does have the required narrow-focus interpretation. This is due
to the fact that overseas conventionally denotes a place or goal, not a path.
References
Bierwisch, Manfred
1988
"On the grammar of local prepositions", in: M. Bierwisch W. Mtsch
I. Zimmermann (eds.), Syntax, Semantik und Lexikon. (Studia grammatica
XXIX.) Berlin: Akademie-Verlag, 1-65.
Brugman, Claudia
1988
The story of over: Polysemy, semantics, and the structure of the lexicon. New
York: Garland.
Cienki, Alan
1989
Spatial cognition and the semantics of prepositions in English, Polish, and
Russian. Munich: Otto Sagner.
Dirven, Rene
1989
"Space prepositions", in: R. Dirven R. Geiger (eds.), A user's grammar of
English, Part B. Frankfurt: Peter Lang, 519-550.
174
John R. Taylor
Fauconnier, Gilles
1985
Mental spaces: Aspects of meaning construction in natural language. Cambridge, MA: MIT Press.
Geeraerts, Dirk
1992
"The semantic structure of Dutch over", Leuvense Bijdragen 81: 205-230.
Grice, H. P.
1975
"Logic and conversation", in: P. Cole J. Morgan (eds.), Syntax and
Semantics 3: Speech Acts. New York: Academic Press, 41-58.
Hawkins, Bruce
1984
The semantics of English spatial prepositions. [Unpublished doctoral dissertation, University of California at San Diego.]
1986
"The preposition out: A case of semantic elision", Linguistic Agency of the
University of Duisburg, Paper A 169.
Herweg, Michael
1989
"Anstze zu einer semantischen Beschreibung topologischer Prpositionen",
in: Ch. Habel . Herweg . Rehkmpfer (eds.), Raumkonzepte in
Verstehungsprozessen. Tbingen: Niemeyer, 99-127.
Jackendoff, Ray
1973
"The base rules for prepositional phrases", in: S. Anderson P. Kiparsky
(eds.), A festschrift for Morris Halle. New York: Holt, Rinehart and Winston,
345-356.
1977
X-bar syntax: A study of phrase structure. Cambridge, MA: MIT Press.
1983
Semantics and cognition. Cambridge, MA: MIT Press.
Jespersen, Otto
1924
The philosophy of grammar. London: George Allen & Unwin.
Johnson, Mark
1987
The body in the mind: The bodily basis of meaning, imagination, and reason.
Lakoff, George
1987
Chicago: Chicago University Press.
Langacker, Ronald
1987
Foundations of cognitive grammar, Vol. 1, Stanford: Stanford University Press.
1988
"A view of linguistic semantics", in: B. Rudzka-Ostyn (ed.), Topics in cognitive
linguistics. Amsterdam: John Benjamins, 49-90.
Radden, Gnter
1989
"Figurative use of prepositions", in: R. Dirven R. Geiger (eds.), A User's
Grammar of English, Part B. Frankfurt: Peter Lang, 551-576.
Schulze, Rainer
1987
"The perception of space and the function of topological prepositions in
English: A contribution to cognitive grammar", in: W. Lrscher R. Schulze
(eds.), Perspectives on language in performance, Vol. 1. Tbingen: Gunter
Narr, 299-322.
1989
"Getting round to (a)round: Towards the description and analysis of a 'spatial'
preposition", Linguistic Agency of the University of Duisburg, Paper A 267.
Taylor, John
1991
"Things, places, and directions", Cognitive Linguistics 2: 357-360.

1992
175
"How many meanings does a word have?", Stellenbosch Papers in Linguistics

25: 133-168.
Vandeloise, Claude
1984
Description of space in French. [Doctoral dissertation, University of California at San Diego.]
Vendler, Zeno
1967
Linguistics in philosophy. Ithaca: Cornell University Press.
Wunderlich, Dieter
1991
"How do prepositional phrases fit into compositional syntax and semantics?",
Linguistics 27: 591-621.
2. Semantic categorization of
prepositions and context
Prepositions and object concepts:

A contribution to cognitive semantics
Priska-Monika
Hottenroth
1. Introduction
There seem to be three basic problems for the semantics of prepositions:
a) Prepositions offer a considerable variability of context-dependent interpretations, i. e., the individual preposition can express a wide range of
relations, and this is true even if the analysis is restricted to the spatial
uses of the individual preposition. The heterogenity of possible interpretations for the individual preposition poses serious problems for all
"abstractionist" theories of word meaning, that is, for instance, for
theories that start from the assumption that word meanings should be
defined in terms of necessary and sufficient conditions for their adequate
use, or that they should comprise all and only information common to
the whole range of context-dependent uses.
b) The second problem for a semantic theory of prepositions arises from
the fact that there is a very close semantic interaction between the
preposition and its linguistic context, especially between the preposition
and its arguments. One cannot think of the meaning of a spatial preposition, for instance, without having some place or thing in mind. Thus,
the meaning of a preposition cannot be conceived without reference to
at least the internal argument, and some configurational information
concerning the arguments seems to be part of the meaning of spatial
prepositions.
Therefore, one of the problems prepositional semantics has to cope
with, is to disentangle and isolate the specific semantic content of the
individual preposition from the contextual factors influencing its various
interpretations, and to make explicit in which way these factors interact
with the prepositional meaning.
c) A third problem for prepositional semantics arises from the fact that
one and the same spatial situation can be viewed from different perspectives, i.e., it can be conceptualized in different ways. In other words, we
are able to create different mental structures to interpret an extralinguistic
scene and, correspondingly, in a number of cases different prepositions
180
can be used to refer to one and the same extra-linguistic situation. Wellknown examples of this phenomenon are the following constructions:
(1)
qqn. dans le champ vs. qqn. sur le champ

'someone in the field' vs. 'someone on the field'
(2)
les pay sans dans I'tte vs. les pay sans sur 1'ile
'the farmers in the island' vs. 'the farmers on the island'
(3)
qqc. dans I'herbe vs. qqc. sur I'herbe

'something in the grass' vs. 'something on the grass'
(4)
qqn. dans la plaine vs. qqn. sur la plaine

'someone in the plain' vs. 'someone on the plain'
(5)
qqc. dans I'assiette vs. qqc. sur I'assiette

'something in the plate' vs. 'something on the plate'
and many others. 1

This means that conditions on the use of the individual preposition
cannot reliably be extrapolated from the objective geometrical properties
of the extralinguistic situations which the preposition can express. Instead,
we have to look for the conditions and cognitive principles that make it
possible to conceptualize certain extralinguistic situations in one way or
the other, and thus to choose one preposition or another.
I assume that a semantic theory for spatial prepositions should be
capable of accounting for all three phenomena mentioned so far in an
adequate and natural way. In particular, it should be able to explain the
whole range of context-dependent meanings.
I further assume that the elaboration of such a semantic theory would
require the development of a hypothesis about the semantic content of
the preposition, and that this hypothesis has to be complemented and
justified by a systematic and explicit elaboration of the conditions which
produce the various context-dependent interpretations. This should, eventually, result in a further set of hypotheses of how the semantics of the
preposition is processed, how it interacts with different contextual factors
so that the various context-dependent interpretations come about. If we
accept that a preposition expresses a relation between two arguments,
these arguments and especially the inner argument, as will be shown
below will naturally play the central part in these processes.
It is at this point that the spatial properties of the argument objects,
i. e., the spatial properties of the located object and the reference object,
have to be taken into account, and with them the principles of object
Prepositions and object concepts
181
conceptualization and categorization relevant for the use of the individual

preposition and its interpretation.
It is one of the main purposes of this paper to try to show in which
way the different properties of the argument objects, as well as certain
principles of object conceptualization and categorization, and the systematic mental manipulation of object concepts participate in bringing about
the diversity of possible interpretations of spatial prepositions.
On a more general level, the aim is to show that certain kinds of
"polysemies" for which the polysemies of spatial prepositions are but
an example can be explained in a very natural and psychologically
plausible way by taking these principles into account.
Until recently, linguists very decidedly tended to consider questions
like object conceptualization and categorization to be outside the realm
of linguistic research and not part of the subject matter of lexical semantics. This seems to have changed with the cognitive linguistic perspective
which explicitly opens up a way of accounting for the connection between
the usage of linguistic expressions and the way the world is perceived
and conceptualized. And, indeed, not much progress seems to be possible
otherwise, not only in the semantics of spatial expressions, but also in
other domains.
In what follows I shall analyze the French preposition dans. The
analysis will be restricted to the spatial uses of this preposition. The aim
of this paper is, as has already been said, not only to provide a semantic
analysis of dans, but rather to propose a framework for semantic analysis
which could easily be generalized to other spatial prepositions and spatial
expressions in general, and which could, eventually, even be extended to
linguistic expressions in other domains. Of course, such an analysis would
bring up a host of further general cognitive principles and principles of
concept manipulation beyond those involved in the use and interpretation
of dans, as they are reconstructed in this paper.
2. General assumptions
I shall start from the following general assumptions in lexical semantics:
1) There are two cognitive levels involved in the interpretation of a
linguistic expression: a level of semantic structure, where the linguistically
relevant knowledge associated with the lexical item is coded, and a level
of conceptual structure where our world knowledge is organized. The
182
structures on the second level, determine the ways in which our experience
of the world is conceptualized and interpreted. 2
2) The semantic structure of a lexical unit must contain information
which determines its appropriate use in the process of language production and constrains its possible interpretations in language understanding.
3) Interpretations of linguistic expressions belong to the conceptual level.
They are the result of processing the information coded on the semantic
level, with the related conceptual and contextual knowledge.
4) It is assumed that these interpretation processes are guided and determined by general cognitive principles.
Although I differentiate between a semantic and a conceptual level,
this should not imply a commitment on my part to the following very
fundamental theoretical question in recent discussion: Is the semantic
structure basically different in nature from conceptual structure? (As
assumed in the framework proposed by Bierwisch and Lang (1987).3 Or,
are semantic structures a "subset of conceptual structures just those
that happen to be verbally expressible"? (As assumed by Jackendoff
1983.4
In fact, I think that the results of the analysis I shall present will not
be affected greatly by the choice of one or the other of these models, and
could be integrated into either. What I am mainly concerned with are
the questions: What is the kind of information that is in one way or
other directly associated with the lexical unit or which we have access to
via the lexical unit? (In other words what are the concepts (or conceptual
structures) we need for their interpetation?) And which are the principles
processing these concepts with different contextual information so that
the various context-dependent interpretations come about? In either semantic model this information (or these conceptual structures) have to
be reconstructed from the analysis of the range of uses of the individual
lexical item, and this hypothesis has to be justified by elaborating the
principles of their processing in a way that would explain the whole range
of their use.
Thus, the question of whether this conceptual information is itself
coded in the semantic representation or if the information coded there is
of a principally different kind, (but constrains the possible instantiations
on the conceptual level) can be left undecided for the moment. It is only
for convenience that, in the following pages, I shall talk of this information as: "the information coded in the semantic representation".
Within the framework sketched so far, I shall try to reconstruct and
specify informally which kind of (conceptual) information must be coded
183
in the semantic representation of the preposition dans to constrain the

set of possible interpretations, and which kind of processing principles
are needed so that the whole range of possible context-dependent interpetations can be accounted for.
I want to stress one more point which will be of some importance for
the following analysis:
I assume that any hypothesis regarding the conceptual information coded
in the semantic representation of a lexical unit can be taken for justified
if and only if principles of processing can be found which are general
cognitive principles in the sense that they also operate in other cognitive
domains.
3. General structure of the semantics of prepositions and

first general assumptions about the interaction with the
context
Considerable work has already been done on the problem of the general
semantic structure to be assumed for spatial prepositions. I shall follow
a proposal made by D. Wunderlich (1982)5 in that I assume that the
meaning of a spatial preposition is a relation of localization between the
localized object and a characteristic region with respect to the reference
object y.
Following Wunderlich (1982), . Herweg (1988) proposed the following semantic representation:
(6)
y y (LOK (, PREP* (y)))6
This assumption is based on the observation that the meanings of prepositions like dans, sur, sous, devant, derriere, and others, differ from each
other in that each preposition relates the localized object to a different
characteristic region with respect to the reference object. (See Figure 1.)
If we adopt this proposal this means that the specific semantic contribution of each preposition lies in the component PREP*. In the case of
dans, this component will be called DANS* in the following. Thus,
DANS* must contain the information that determines the kind of regions
dans can identify with respect to different possible reference objects. In
other words, the component DANS* must contain the abstract conditions
for the identification of all and only the possible i/am-regions.
184
Figure 1. the regions the different prepositions select on a three-dimensional object
The processing of the component DANS* with its various reference

objects, (more precisely the concepts of its reference objects7), will therefore have to be considered the central part of the contextual interpetation
of the preposition. The result of this process is one or more conceptual
instantiations of that component, i.e., in the case of dans, the identification of one or more possible ifa.y-regions with respect to the reference
object in question. Compare for instance the normal interpretations of:
(7)
les peches dans le bol

'the peaches in the bowl'
and
(8)
la fissure dans le bol

'the crack in the bowl'
(See Figure 2.)

(9)
les objets dans la boite

'the objects in the box'
(10)
le trou dans la boite

'a hole in the box'
The interaction of dans with boite, for instance, yields two possible regions
with respect to the reference object: the (hollow) region enclosed by the
185
Figure 2.
box in (9) and the material of the box, i. e., the region materially occupied
by the reference object in (10).
Other contextual factors that have an influence on the interpetation of
the preposition seem to operate only on the result of this identification
procedure. Thus for instance, in the examples above, the properties of
the located object (together with general knowledge about the normal
spatial relations between the located object and the reference object)
determine the selection of the region meant among the possible dansregions identified.
Compare also the normal interpretations of the following examples:
(11)
I'oiseau dans I'arbre

(12)
la hachette dans I'arbre

'the axe in the tree'
(See Figure 3.)

Contextual knowledge other than the properties of the located objects
determines the selection of the region meant (among those identified) in
ambiguous constructions like:
(13)
I'eclat de verre dans la main

'the glass splinter in the hand'
(14)
le clou dans la caisse

'the nail in the box'
(See Figure 4.)

Quite another type of interpretation difference can be observed:
186
l'oiseau dans l'arbre

Figure 3.
(15)
la pipe dans la bouche

'the pipe in the mouth'
(16)
le bonbon dans la bouche

'the sweet in the mouth'
(17)
le paquebot dans I'eau

'the steamboat in the water'
(18)
les poissons dans I'eau

'the fish in the water'
(19)
les fleurs dans le vase

'the flowers in the vase'
(20)
I'eau dans le vase

'the water in the vase'
(See Figure 5.)
Figure 4. le clou dans la caisse
les fleurs dans le vase
187
l'eau dans le vase
Figure 5.
The difference in interpretation is that between total and partial inclusion.

But the region that dans identifies is the same within each of the pairs of
examples. And it is the characteristic geometrical properties of this region
that determine the use of dans and not sur, sous or any other preposition
in each of these examples, whereas the interpretation difference between
total and partial inclusion is determined by general knowledge about the
usual spatial relations between the located object and the reference object.
The same is true for interpretation differences such as whether the
located object is compact, i.e., remains a separate object inside the
reference object, or whether it mixes with it, desolves in it or even reacts
chemically with it as for instance in:
(21)
la cuillere dans le cafe

'the spoon in the coffee'
(22)
le sucre dans le cafe

'the sugar in the coffee'
(23)
le lait dans le cafe

'the milk in the coffee'
188
(24)
l'oxygene dans l'eau

'the oxygen in the water'
Thus, the interaction of dans with the reference object seems to come
first in the interpretation process, and it is only then that secondary
processes operating on the result of that interaction select one of the
possible dans-regions identified, or determine differences in interpretations such as total and partial inclusion, and others. Therefore the central
and linguistically most interesting part of the analysis of the different
possible interpretations of dans will concern the component DANS* and
its interaction with the different possible reference objects. Our analysis
will be restricted to this part and thus to the following questions:
a) What sort of information must be coded in the central component
DANS* so that all and only the possible dans-regions are identified with
respect to the different reference objects?
b) Which are the principles that guide this identification process?
The point of interest with respect to the second question lies in the fact
that due to the very heterogeneous spatial properties of possible reference
objects, the geometrical properties of the possible i/aws-regions are very
diverse, as will be shown in detail in the next chapter.
4. The diversity of possible dans-regions

Generally it is assumed that the information coded in the semantic
representation should be a generalization covering all possible instantiations on the conceptual level. All attempts at specifying DANS* by
formulating such a general abstraction which should cover all possible
conceptual instantiations have met with considerable difficulties. This is
because, as has been said, instantiations of DANS*, i. e., regions dans
can identify, can have very diverse geometrical properties. See the following list (which is not at all complete, but in which the examples have
already been tentatively ordered):
(25)
la lettre dans la boite

'the letter in the letter-box'
(26)
le vin dans la bouteille

'the wine in the bottle'
(27)
le vin dans le verre

'the wine in the glass'
(28)
la viande dans I'assiette

(also: la viande sur I'assiette)8
'the meat on the plate'
(29)
un vieil homme dans son fauteuil

'the old man in the armchair'
(30)
le bebe dans les bras de son pere

'the baby in his father's arms'
(31)
I'homme dans l'encadrement de la porte

'the man in the doorframe'
(32)
la grosse tabatiere dans ses doigts

(also: entre ses doigts)9
'the big tobacco-box held by his fingers'
(33)
I'homme dans la foule

'the man in the crowd'
(34)

(35)
le clou dans la planche

(36)

(37)
un trou dans le papier

'a hole in the sheet of paper'
(38)
le ballon dans I'air

'the balloon in the air'
(39)
la vie dans I'espace

'life in space'
(40)
le monastere dans l'enceinte de fosses

'the monastery enclosed by the moat'
(41)
les maisons dans cette region

'the houses in this region'
189
190
(42)
la ville dans ses murailles

'the town within its walls'
(43)
l'homme dans le cercle

'the man in the circle'
In view of the heterogenity of these and other possible i/ans-regions, how

could DANS* be specified in order to insure the identification of all and
only the possible dans-regions? Is it at all possible to define a set of
properties that are common to all of them?
Before I go deeper into this question, and in order to give a somewhat
clearer picture of the range of possible ifaws-regions and the relations
among them, a classification will be presented in the following. The
criteria for this classification will be those obviously relevant for the use
of dans as opposed to sur, sous etc., i. e., the dimensionality of the region,
its boundedness or unboundedness, and whether or not it is materially
occupied. The different classes of interpretations of dans will be called
use-types.10 They differ from one another in that the region dans identifies
has different properties as regards dimensionality, boundedness or being
materially occupied.
5. A systematic classification of the possible daws-regions11

The systematic analysis of the whole range of possible uses of dans as we
found them in our corpus, and their classification according to the
different relevant geometrical properties of the i&zs-regions has yielded
the following use-types:
Use-type 1: The dans-region (the region dans is related to) is threedimensional, hollow (i. e., not materially occupied by the reference object)
and materially bounded on all sides.
We have this kind of region in the normal interpretations of the
constructions already mentioned:
(44)
les objets dans la botie

'the things in the box'
(See Figure 6.)

(45)
les robes dans I'armoire

'the clothes in the wardrobe'
191
Figure 6. les objets dans la boite
(46)

and in all similar constructions where the reference object is a closed

container, and a container/contained relationship is to be expressed. This
seems to be the most frequent and most typical use of dans. It is the one
that comes to mind first and the one which obviously is learned first in
the child's language acquisition. This fact will be taken up again, and
commented upon in the next chapter.
We have this kind of region not only with reference objects that are
containers, but also with houses, cars and other vehicles, hollow organs
etc. as for instance in:
(47)
les enfants dans la maison

'the children in the house'
{AT)
le sang dans le caeur

'the blood in the heart'
(48)
le liquide dans I'estomac

'the liquid in the stomach'
(49)
le bonbon ou la pipe dans la bouche

'the sweet or the pipe in the mouth'
and many others.

With all these reference objects the hollow region is a constitutive part
of the reference object as determined by its category. But this need not
be the case, as is shown by the following examples:
(50)
les fleurs ou le cadeau dans le papier jaune

'the flowers or the present in the yellow paper'
(51)
les jetons dans la main

'the chips in the hand'
192
With all these and similar examples, the hollow but materially surrounded
dans-region is induced by dans (together with knowledge about the typical
relations with the located objects). The prerequisite for this mechanism
is that the reference objects be malleable objects, i. e., objects that can be
given a form so that they can enclose something.
Other examples where the region dans identifies has the properties defined
for use-type 1 are:
(52)
la lettre dans le journal

'the letter in the newspaper' (in the sense of lying between the
pages of the newspaper) 12
(53)
les fleurs sechees dans le livre

'the dried flowers in the book'
(54)
la noix dans sa coquille

'the nut in its shell'
(55)
la peche dans sa peau

'the peach in its skin'
and many others. As can be seen, the examples belonging to use-type 1

are rather diverse if one thinks in terms of relation differences and
differences of object categories ivolved. Yet, in each of these examples
(in their normal interpretation or at least in one of their possible interpretations) the dans-region has the properties defined in use-type 1: it is
three-dimensional, hollow, and materially enclosed and thus bounded by
the reference object.
Use-type 2: The dans-region is three-dimensional, hollow (in the sense of
not being materially occupied by the reference object), but only partially
bounded by the material of the reference object.
This type of region is meant in the normal interpretations of examples
like:
(56)

'the peaches in the bowl'
(See Figure 7.)

(57)
le vin dans le verre

'the wine in the glass'
(See also Figure 8.)

(58)
I'eau chaude dans la baignoire

'the warm water in the bathtub'
193
la banane dans le bol
Figure 7.
Strictly speaking, and from a material point of view, the located object
is outside the boundaries of the reference object. This is all the more
obvious the less closed and the more open the concavity is. (See Figure 9.)
The possibility of using dans seems to presuppose that the region partially
enclosed by the material of the reference object be mentally closed and
that it be conceptually treated as part of the reference object. Most
probably, the mental completion of the material boundary of the dansregion can usually be thought of as a horizontal plane as in Figure 8.
But compare again Figure 7, where a banana is placed on top of some
other fruit in a bowl in such a way that the mental completion of the
material boundaries of the dans-region by a horizontal plane would still
leave the banana outside this region. The relation would still be expressed
by dans:
(59)
la banane dans le bol

'the banana in the bowl'
As dans can still be used in these cases the kind of mental closure to
be assumed seems to be functionally motivated and can thus go beyond
the simple horizontal closing plane (see the curved dotted line in Figure
7). The mental manipulation of closing the material boundaries of the
Figure 8. la mouche dans le verre
Figure 9. la viande dans l'assiette
194
dans-region seems to be more or less automatically given with the categories of open containers, open vehicles etc., and the respective region
is constitutive of these categories. In principle, all categories of objects
that can figure as reference objects in use-type 1 can also appear as
reference objects in this use-type, because for instance containers (or
similar objects providing a hollow region) can be closed or open (or in
any way not quite closed) so that the region identified by dans is not
totally enclosed materially. Thus, for instance:
(60)
les robes dans la valise

'the dresses in the suitcase'
(61)

(62)
les robes dans l'armoire

'the dresses in the wardrobe'
would belong to this use-type in case the containers are not closed and,
similarly, also:
(63)
les jetons dans la main

'the chips in the hand'
(64)
la lettre dans le journal

'the letter in the newspaper' (in the sense of: between the pages
of the newspaper)13
(65)
les fleurs sechees dans le livre

'the dried flowers in the book'
if the hand, the journal and the book are half open. In all these cases the
reference objects would not enclose the dans-region on all sides.
But regions of this kind are also offered in examples like.
(66)
la femme dans la petite barque

'the woman in the little boat'
(67)
un homme dans son fauteuil

'the man in the armchair'
(68)
l'ampoule dans la douille

'the bulb in the socket'
(See Figure 10.)
195
Figure 10. l'ampoule dans la douille
(69)
la mouche dans la lampe

'the fly in the lamp'
(70)
I'eau dans le trou

'the water in the hole'
and many others.

Within use-type 2 dans-regions exhibit various continua of partial
(material) boundedness:
Containers of all kinds can be more and more open; they can even be
flat plates:
(71)

'the meat on (in) the plate'
There are reference objects which parially enclose regions that are open
on more than one side. This is the case with tubes, trenches, furrows,
grooves, cages, sieves etc. See the normal interpretations of:
(72)
le gaz dans le tuyeau

'the gas in the tube'
(73)
I'eau dans la tranchee

'the water in the ditch'
and also of:

(74)
les voitures dans la rue

'the cars in the street'
and many others.
196
As is the case in constructions with open containers as reference objects,

the prerequisite for the use of dans is that the region partly enclosed be
mentally closed and treated as part of the reference object.
The material boundary of the region identified can be reduced to a
ring or even to some points only, as in:
(75)
la serviette dans I'anneau

'the napkin in the ring'
(76)
la vis dans I'ecrou

'the bolt in the nut'
(77)
I'homme dans l'encadrement de la porte

'the man in the door frame'
(78)
la tabatiere dans ses doigts

'the tobacco-box held by his fingers'
(79)
I'insecte dans les pincettes

'the insect gripped by the tweezers' 14
(See also Figure 11.)
Figure 11. I'ecrou dans les tenailles
The more the material boundary of the region dans identifies is reduced
in these cases, the more the mental closure of the region, and thus the
possibility of using dans, seem to be dependent on the function of the
reference object as a container, or at least on a kind of holding function.
Thus the mental closure has to be functionally motivated in these cases.
The mental completion of the material boundary seems also to be the
condition for the use of dans in the following examples. This closure
seems rather to be motivated visually, not functionally in these cases:
197
(80)

'the bird in the tree' (in the sense of "in the tree-top between
the branches and leaves")
(81)

(See Figure 12.)

0
g0#00 :
Figure 12. I'homme dans la foule
Dans can only be used here if the reference object is conceptualized as

an entity, not as a collection of individual objects. Otherwise, for instance
with the plural personnes in (81) or branches in (80) entre would have to
be used instead of dans.
Again within this use-type the differences among the examples classified
here are enormous, if we think in terms of differences of object categories
involved and also of relational differences. Yet again, they have in
common the kind of region dans indentifies. In all the examples mentioned
assuming their normal interpretations (or, at least, one of the possible
interpretations) the dans-region is three-dimensional, hollow, and only
partially bounded by the material of the reference object.
Use-type 3: The dans-region is three-dimensional, (normally) occupied by
the material of the reference object, and bounded on all sides.
We have i/aws-regions of this type in:
(82)

(83)
le couteau dans le pain

'the knife in the bread'
198
(84)
la hachette dans le tronc d'arbre

'the axe in the tree-trunk'
(See Figure 13.)
Figure 13. la hachette dans le tronc d'arbre
(85)
le trou dans le fromage

'the hole in the cheese'
and many similar constructions, but also in:

(86)

(87)
le trou dans I'armoire

'the hole in the cupboard'
and many others.

Examples where the reference object is a quantity of a substance also
belong to this use-type, although very often the boundaries of the dansregions seem to be out of focus in the respective interpretations.
(88)
le clou dans le bois

'the nail in the wood'
(89)
le sucre dans le cafe

'the sugar in the coffee'
(90)
le ballon dans I'air

'the balloon in the air'
(91)
le pied dans la neige

'the foot in the snow'
(92)
la main dans I'eau

'the hand in the water'
199
and similar constructions. In (91) and (92), the upper surface is one
relevant boundary of the dans-region in question. All other boundaries
seem to be more or less out of focus in most of the constructions of this
use-type. Reference objects of dans can also be substances in general,
instead of limited quantities of a substance. This use-type seems to be
simply extended to such cases. 15
Use-type 4: The dans-region is three-dimensional and vaguely bounded
or unbounded.
Totally unbounded are the Jaws-regions in the following examples:
(93)
les galaxies dans I'espace

'the galaxies in space'
(94)
les cometes dans le voisinage de la planete

'the comets in the neighbourhood of the planet'
(95)
la vie dans I'univers

'life in the universe'
and others.
It seems, however, that on the conceptual level some kind of boundary
is mentally created in these cases. Yet the mechanism seems to be different
from the mental completion of partial boundaries as with open containers
and similar objects: For instance, the mentally imposed boundaries are
not available in the object concepts in a way that would make the use of
hors de, dedans, entrer, or sortir admissible.
Use-type 5: The dans-region is two-dimensional and bounded.
Regions dans identifies do not only differ with respect to their being
materially occupied or hollow, or with respect to their being materially
bounded, partially bounded or unbounded, but they can also differ as to
their dimensionality. They are not only three-dimensional, as in all instances mentioned so far, but also two- and even one-dimensional (See
below use-type 8). They are two-dimensional in the following examples:
(96)
le point dans le cercle

'the point in the circle'
(97)
la boule dans le carre

'the ball in the square'
(98)
les pyldnes dans ce carre

'the pylons in that square field'
(See Figure 14.)
200
Use-type 6: The dans-region is three-dimensional, and bounded, but only

two dimensions are relevant for the use of dans.
This is the type of region we have in constructions where the reference
objects are geographical regions, territories etc. It is a frequent and very
interesting group of uses. See for instance:
(99)
les usines dans la France d'aujourd'hui16

'the factories in contemporary France'
(100)
les caravanes dans le desert

'the caravans in the desert'
(101)
les saboteurs dans ce me me territoire

'the saboteurs in this same territory'
(102)
les lapins dans le champ

'the rabbits in the field'
(103)
les moutons dans le pre

'the sheep in the meadow'
(104)
les fleurs dans le jar din

'the flowers in the garden'
(105)
les animaux dans la foret

'the animals in the wood'
(106)
les chalets dans les Alpes

'the cottage in the Alps'
and similar constructions.

In all these cases the reference objects are three-dimensional and thus
offer three-dimensional regions that dans could be related to, but only
201
two dimensions seem to be relevant for the localization relation expressed

by dans. It is really only the localization with respect to the horizontal
plane which is expressed. It is territorial delimitation that motivates the
use of dans, not any elevations above the ground or any mental closure
of a three-dimensional region above the ground, as has sometimes been
suggested. The dans-region, in all these cases, is a region which is delimited
as a part of a larger region of the same kind. The same type of region is
identified in constructions where the reference object is a delimitation or
a bounding object like a fence or wall, enclosing a territory.
(107)
le monastere dans l'enceinte de fosses

'the monastery encircled by the moat'
(108)
la ville dans ses murallies

(109)
les mountons dans le cloture

'the sheep within the fence'
Use-type 7: The dans-region is two-dimensional, or only two dimensions

are relevant, and it is partially bounded, vaguely bounded, or unbounded.
Both geometrical two-dimensional entities and geographical regions can
be partially bounded, vaguely bounded, or even unbounded. Consider,
for instance:
(110)
le point dans I'angle

'the point in the angle'
(See Figure 15.)

(111)
les maisons dans le voisinage

'the houses in the neighbourhood'
Figure 15. le point dans Tangle
202
(112)
cette petite tente dans l'immense et endue

'the little tent in the immense vastness'

The prerequisites for using dans in these cases seem to be the same
kind of mental manipulations of the object concept as mentioned under
use-type 2 and use-type 4: the boundaries must be completed mentally
so that they enclose the region to be identified by dans as in (110), or
boundaries are created mentally as in (111), or left out of focus as in
(112).
Use-type 8: The dans-region is one-dimensional.
See, for instance:
(113)
intervalle dans la ligne

'the break in the line'
(114)
I'astigmatisme dans I'axe

'the astigmatism in the axis'
and very few similar cases.

This is a rather marginal use-type. Examples are rare and not all of
our French informants accepted them. But some did and we found them
in our corpus, so they are produced and understood. This is why we
include this use-type here. There seem to be very few nouns referring to
one-dimensional objects in French, but also the class of located objects
that can be related by dans with one-dimensional reference objects seems
to be rather limited. See for instance the marginality of the following
constructions:
(115)
le point dans la ligne

'the point in/on the line'
and the unacceptibility of:

(116)
*la pyldne dans la ligne

'the pylon on the line'
In both cases the relation can only be expressed by sur. All this shows
that we have reached the borderline of the applicability of dans in this
use-type.
To summarize: the analysis has shown that the possible ifons-regions
can be three-dimensional, two-dimensional and even one-dimensional.
Dans-regions can be bounded, vaguely bounded, partially bounded, or
unbounded, and they can be hollow or materially occupied by the ref-
203
erence object. If the dans-region is a hollow region it can be a constitutive

part of the reference object (as determined by its category) or it can be
a region which is only surrounded by the reference object, but not part
of it.
6. The difficulty of specifying *DANS

In view of the diversity of the possible dans-regions the question is again:
How is the semantic component DANS* to be specified? Suppose we
assume that it should be specifiable in some way, that it should embody
the conditions of identification for all and only the possible dans-regions,
and that the information coded in this component should be a generalization over the whole range of the different possible dans-regions: then
the problem is that neither dimensionality, nor boundedness, nor being
hollow or materially occupied by the reference object are properties that
qualify, because none of them is given in all uses. And this is so even
though these are exactly the properties which, at least in the great majority
of cases, determine the use of dans as opposed to sur, sous, entre, or
certain other prepositions.
The specification of DANS* by global concepts like PLACE (Y)17 or
INTERIOR (y) cannot really solve the problem. For instance, if we take
INTERIOR (y) as a generalization over all kinds of possible dans-regions,
this concept would have to cover not only regions as offered by reference
objects like open containers of all kinds, but also those offered by plates,
streets, fingers, hands, arms, tweezers etc., as well as unbounded regions,
and two- and one-dimensional regions. If it has to cover all of these, the
concept INTERIOR would have to be too abstract to serve as identification criteria. It would be too abstract to account for the oppositional
relations of the preposition or for the restrictions of use. For instance, it
would not explain the difference of acceptability between:
(117)
l'Intervalle dans la ligne

and:
(118)
lie point dans la ligne

?'the point in the line'
nor the unacceptability of:
204
(119)
*qqc. dans le point

*'something in the point'
If we take PLACE (y) as the specification of the component DANS*, we

shall meet with much the same problems. In particular, we shall have
problems with examples where the region identified by dans is not the
place the reference object itself occupies, or where hollow Jaws-regions
are not a constitutive part of the reference object. See, for instance,
(120)
le cadeau dans le papier jaune

'the present (wrapped) in the yellow paper'
(121)
le monastere dans l'enceinte de fossees

'the monastery within the moats'
(122)
la ville dans ses murallies

(123)
le corps dans le linceuil

'the body in the shroud'
(124)
la tabatiere dans ses doigts

'the tabacco-box held by his fingers'
(125)
I'ecrou dans les tenailles

'the screw held by the tongs'
and many others.

The specification PLACE (y) would also not explain the marginality
of cases like:
(126)
Tie point dans la ligne
nor the restriction prohibiting 0-dimensional objects as reference objects

of dans:
(127)
*qqc. dans le point
where only sur would be possible.

Neither INTERIOR (y) nor PLACE (y) would be capable of explaining
why dans ans sur express a clear opposition in extralinguistic relations
in:
(128)
versus:
qqc. dans la surface

'something in the surface'
(129)
205
qqc. sur la surface

'something on the surface'
while the extralinguistic situation expressed is the same in the normal

interpretations of:
(130)
les tentes dans la plaine

'the tents in/on the plain'
(131)
les tentes sur la plaine

'the tents on the plain'
as well as in:
(132)
les pay sans dans I'lie

'the farmers in/on the island'
and:
(133)
les pay sans sur I'ile

'the farmers on the island'

Thus the assumption of a global concept like those mentioned would
mean giving up a specification of DANS*, and thus an explicit specification of the identification criteria for possible ifans-regions. This would
mean that the crucial part of the semantics of the preposition would be
left out of consideration.
Yet, in view of these difficulties, the question indeed arises if it is at
all possible to specify DANS* in a way that the coded identification
criteria for possible Jans-regions become explicit, and whether this specification can be a generalization over all possible instantiations?
7. The prototype approach and its problems

Prototype semantics seems to offer a solution to just this kind of problems
by introducing the theory of natural categorization, developed by E.
Rosch (1975, 1977) and others, into lexical semantics, using it as a model
to account for the phenomenon of "polysemy" of linguistic expressions.
The different, but intuitively interrelated senses of a word are thus treated
in analogy to the different individual members subsumed under a natural
category.
206
Just as the theory of natural categorization considers it psychologically

inadequate and superfluous to try to define the category by necessary
and sufficient conditions for membership (or by properties common to
all members of the category), the prototype approach in lexical semantics
no longer aims at a definition of word meaning that would be an
abstraction from all possible uses of the word. Instead, it is assumed that
word meanings are much more specific; that "polysemous" words have
a prototypical core meaning that covers only the most typical instances
or is an abstraction only from the central, most typical and most frequent
range of uses. Apart from uses with properties corresponding to those
defined in the prototypical core meaning, the model allows for deviations
in meaning of various kinds, i.e., meanings which would not have all the
properties defined in the core meaning, but which would show a "family
resemblance". Just as in a natural category members that deviate from
the prototype are subsumed on the basis of similarity judgments, according to this theory similarity judgments are also at the basis of polysemy;
polysemy comes about by extending the use of a word gradually to
extralinguistic phenomena which do not have all of the properties defined
in the word meaning and thus do not correspond to the prototypical
cases but are judged to be similar.
This kind of prototype approach has been applied to the semantics of
spatial prepositions by quite a few linguists, for instance A. Herskovits
(1986); B.W. Hawkins (1984); Vandeloise (1986) and also Hottenroth
(1986). Within these works, there was a growing interest in the fact that
deviations from the prototypical core meanings defined for the individual
prepositions did not seem to be arbitrary, but systematic. The aim was
to make these regularities explicit, and some even tried to translate them
into corresponding principles by which deviating uses of an expression
could be generated starting from the prototypical core. This would give
a certain predictive force to the semantic model. (See especially Vandeloise
1986 and Hottenroth 1986.)
Although the prototype approach has certainly yielded the most detailed and exhaustive studies of the polysemies of the spatial prepositions,
it has also been criticized for leaving quite a lot of theoretical and
methodological problems unsolved. Among them, first of all, how can
the prototypical core meaning be determined in a systematic and principled way? By what criteria can we know on which level of abstraction it
should be defined? In principle, different levels of abstraction could be
equally plausible. The core meaning could thus be more or less specific,
and what would be considered as deviations would vary accordingly.
207
How can we tell which of the different resulting analyses is semantically

adequate and psychologically plausible?
Secondly, what is the status of the deviations from the core meaning
within the semantic model? Are they also coded in the lexicon (i. e., is
the semantic model truly polysemous) or are they generated by certain
principles operating on the core meaning? And if so, what sort of status
would these principles have within the theory?
Thirdly, not all of the possible deviations from the core meaning
described within this semantic model seem to be linguistically relevant
deviations. On the contrary, quite a number of them seems to be linguistically rather irrelevant (especially those concerning continua in the extralinguistic world where we seem to have to do with purely referential
polysemy). Yet some deviations obviously are linguistically relevant meaning differences. The prototype model does not seem to account for such
differences in status.
Fourthly, a simple transfer of the model of natural categorization to
lexical semantics does not really seem to be an adequate semantic model,
at least not in the case of relational expressions like prepositions. For, as
the prototype semantic analyses have themselves shown, phenomena in
this domain are more complex than that. It is not simple similarity
judgments associating deviant spatial relations with the prototypical kind
of relation that produce the range of different interpretations of a spatial
preposition. Rather, we must reckon with more sophisticated processes
of mental manipulation of object concepts, of dimension projection and
metaphorical restructuring that make the prepositional meaning and the
different reference objects fit together, as will be shown in the next chapter.
8. The proposal
The semantic model I propose for the analysis of dans (and other spatial
expressions) will deviate from prototype semantics in almost all respects,
above all in that I assume that the different contextual interpretations
are not part of the semantics of the preposition but are the result of
processing the semantic representation of the preposition with its context,
and that they belong to the conceptual level. However, I shall take over
the idea that the semantic information coded with the preposition need
not be an abstraction over all contextual interpretations, but that it is
more specific and that it is an abstraction only from a central range of
208
uses. Thus also the central component DANS* in the semantic representation of the preposition need be an abstraction only from the central
range of possible dans-regions and should be specified accordingly.
I assume that the kind of specification and the level of abstraction for
DANS* is correctly chosen and can be considered as justified if it is
possible to reconstruct principles that determine the processing of DANS*
with the reference object and are general cognitive principles yielding the
whole range of possible dans-regions. Given these conditions, the most
appropriate candidate for the specification of DANS* seems to correspond to the concept of a region that is three-dimensional, hollow and
materially bounded on all sides by the reference object. Defined in this
way DANS* does in fact embody identification criteria valid only for the
central range of uses, namely all examples belonging to use-type 1, as for
instance.
(134)
les objets dans la boite

'the things in the box
In their normal interpretations, all the examples in use-type 1 are direct

instantiations of DANS*, i. e., the dans-regions identified all have the
properties defined in DANS*.
It can now be shown that all other instantiations of DANS*, i.e.,
regions with properties deviating from those coded in DANS*, are identified as possible dans-regions on the basis of a small set of general
principles. These principles either mentally manipulate the deviating region offered by the reference object so as to create the properties required,
or they produce metaphorical or dimensional projections. They are all
principles known from the processes of object conceptualization and
categorization in general.
Thus, regions with the properties characteristic for use-type 2 (i.e.,
three-dimensional regions, which are hollow and only partially bounded
by the material of the reference object) are identified and thus subsumed
as possible dans-regions on the basis of the mental operation of Gestaltschlieung, i. e., the mental closure of the material boundaries. The principle of Gestaltschlieung seems to be one of the most general principles
of object conceptualization and categorization. As has already been said
in the chapter on use-type 2, this is a mental manipulation of the object
concept, which is often, not always, part of the categorization of the
reference object, as is, for instance, the case with open containers:
(135)
la peche dans le bol

'the peach in the bowl'
209
This mental manipulation is the necessary condition for the use of dans
as opposed to sur, or to entre. See also the following examples:
(136)

(137)
I'insecte dans les pincettes

'the insect held by the tweezers'
It has already been mentioned that there is a wide range of different

continua with respect to the reduction of the material boundaries delimiting the Jaws-regions, because reference objects of dans can be bowls
(with different degrees of opening), plates, tubes, rings, screw-nuts, arms
and in French even fingers, tongs, and tweezers. As seems to be the case
with all continua in the extralinguistic world, subsumption on the basis
of similarity judgments seems to play a rather important additional role
in this domain.
Unbounded and vaguely bounded regions, as defined in use-type 4,
are identified as possible /czMs-regions on the basis of a principle of mental
creation of boundaries.
(138)
les galaxies dans I'espace

'the galaxies in space'
The mental creation of boundaries seems to be similar to the principle

of "Gestaltschlieung", but not identical. It seems to be a very basic
principle involved in the processes of perceiving and conceptualizing
entities. Even entities like "space", "universe", "infinity" and the like
apparently cannot be conceived of without some sort of mental delimitation.
Regions which are not hollow, but materially occupied by the reference
object, as for instance in:
(139)

and all other examples in use-type 3 are identifiable as possible dansregions due to a kind of basic metaphorical process creating a structural
analogy between the hollow region bounded by a layer of solid material,
and a piece of solid material bounded by its own outer surfaces.
The identification of two-dimensional and one-dimensional regions as
possible Jam-regions, (i.e., of the regions characteristic of use-type 5 and
8) presupposes the very general principle of projecting three-dimensional
210
configurations onto two dimensions and two-dimensional configurations

onto one dimension. Consider again:
(140)
le point dans le cercle

'the point in the circle'
(141)
I'intervalle dans la ligne

An especially interesting case is the identification and subsumption of

geographical regions as dans-regions (see use-type 6). Here the threedimensionality of the geographical entities is first reduced to the only
two dimensions relevant for the localization relation, i. e., to the horizontal plane. The result is treated as a two-dimensional region, i. e., subsumed
on the basis of a principle of projecting three dimensions onto two.
The combined effect of the principle of dimension projection and the
principle of the mental closure of material boundaries mentally is responsible for the identification of partially bounded two-dimensional
regions (see use-type 7) as in:
(142)

'the point in the angle'
Vaguely bounded or unbounded geographical regions (see also usetype 7) are subsumed by the combined effect of the reduction of the
dimensionality to the two relevant dimensions, the principle of dimension
projection and the mental creation of boundaries, as, for instance, in:
(143)
les maisons dans le voisinage

'the houses in the neighbourhood'
(144)
les maisons dans cette immense etendue

'the houses in that immense vastness'
An overview of the use-types identified and the principles underlying

the identification and subsumption of the different dans-regions is given
in the end.
9. Summary and conclusion

In this paper I started from the following assumptions:
1) The range of different interpretations of lexical units comes about by
processing its semantic representation with the relevant conceptual and
contextual knowledge.
211
2) The core component of the semantic representation of dans is the

component DANS*. DANS* contains the identification criteria for the
possible dans-regions with respect to different reference objects.
3) The central and linguistically most interesting part of the interpretation
process is the processing of DANS* with the different possible reference
objects. It is this latter process which yields the different possible instantiations of DANS*, i. e., the various dam-regions. The diversity of the
possible daws-regions are at the basis of the interpretation differences of
dans. Other interpretation differences result from secondary processes.
4) The central assumption is that the component DANS*, which has to
determine the kind of regions dans can be related to, is not an abstraction
from all possible contextual interpretations, i. e., from all possible dansregions, but is more specific. It is an abstraction only from a prototypical
range of regions and corresponds to the concept of a region which is
three-dimensional, hollow and materially bounded on all sides by the
reference object.
The analysis has shown that this assumption, i. e., the kind of specification and level of abstraction assumed for DANS*, can be considered
as justified by the fact that all possible dans-regions deviating from the
properties coded in DANS* can be accounted for by certain very general
cognitive principles. These are principles that operate on reference objects
or, more precisely, on certain kinds of deviating regions offered by certain
kinds of reference objects. As we have seen, these are principles by which
certain given deviating regions are either mentally manipulated in a way
so that the result corresponds to the properties coded in DANS* (see the
principle of Gestaltschlieung and the principle of mentally creating
boundaries with unbounded or vaguely bounded entities) or they are
basic principles of metaphorical reconstruction or of dimension projection.
Thus, these principles make it possible that certain kinds of regions
with properties different from those defined in DANS* can nevertheless
be systematically identified as possible daws-regions in language production and interpretation. If a reference object does not offer a region with
the properties defined in DANS*, the use and interpretation of dans is
still possible if (and only if) there is a region such that the subsumption
on the basis of the principles elaborated above is possible. This implies
that possible darcs-regions can only deviate from the properties coded in
DANS* in certain specific ways, and it is these principles that determine
in which ways they can deviate. The model therefore has predictive power.
212
It has further been shown that these principles that guide the identifiction and subsumption of deviating dans-regions are general cognitive
principles in the sense that they belong to the cognitive domain of object
conceptualization and categorization. The knowledge of these principles
can be presupposed for all language users; they even seem to be universal
principles.
Note that reference objects are not categorized by dans, as has sometimes been assumed in prepositional semantics. But in my opinion, it is
rather plausible that the same kind of principles as those effective in the
processes of conceptualizing and categorizing objects should also be
exploited in the process of identifying entities like the regions that can
be possible instantiations of DANS*.
The model for the semantics and the interpretation of dans, as far as
we have elaborated it, seems indeed to account in a natural and plausible
way for a number of phenomena which have hitherto presented serious
difficulties for the semantic analysis of dans.
First, the kind of specification assumed for DANS* and the way it is
processed with the various reference objects, on the basis of the principles
elaborated above, explain rather naturally why clear oppositional relations with other spatial prepositions (such as sur, sous, entre and others)
are only found in the realm of the prototypical uses, i.e., in those uses
where the dans-reg\on identified exhibits exactly the properties defined in
DANS*, and not in others, where the dans-region identified is the result
of a mental manipulation.
Secondly, the model accounts for the prototypical structure of the
range of possible ifos-interpretations, i.e., for the fact that there are
more typical, less typical, marginal uses of dans. Uses are more peripheral
the more the given regions deviate from the properties specified in DANS*
and the more their subsumption and identification as a dans-region is
dependent on the effect of the principles. The marginality of:
(145)

'the meat on (in) the plate'
(146)
l'insecte dans les pincettes

'the insect held by the tweezers'
(147)
1'Intervalle dans la ligne

(148)
He point dans la ligne

'the point in/on the line'
213
and the unacceptability (in a concrete, spatial reading) of:

(149)
*qqc. dans le point

*something in the point'
is thus correctly predicated. In one-dimensional reference objects only

very little of the prototypical structure of the dans-region is left, in 0dimensional reference objects there is nothing left.
Thirdly, pairs like those mentioned in the beginning:
(150)
qqn. dans le champ vs. qqn. sur le champ
(151)
qqc. dans I'assiette vs. qqc. sur I'assiette
and similar pairs can also be explained quite naturally within this model.
Reference objects like champ, pre, herbes, He, plaine, assiette and others,
offer regions that do not correspond exactly to the properties specified
in the semantic representation of dans, (i.e., the properties coded in
DANS*) nor to those specified in sur (which would have to be properties
like being bounded form beneath by the uppermost, flat, horizontal
surface of the reference object). But the regions made available by these
reference objects exhibit properties which make it possible either that
they be subsumed as dans-regions on the basis of the principles elaborated
above, or that they be subsumed as possible jwr-regions on the basis of
principles which are, of course, not identical, but similar in nature.
Subsumption as a possible iwr-region presupposes, for instance, the abstraction from all three-dimensional properties of fields and other geographical entities, i. e., from elevations above the ground, elevated boundaries etc. or from elevated rims with plates and similar objects. This is
also a kind of mental manipulation of the given region.
Last, but not least, within this model it becomes more explicit why
some interpretation differences seem to be linguistically irrelevant and
purely referential interpretation differences, while others seem to be
linguistically relevant "polysemies".
Semantically irrelevant are obviously those differences which concern
continua in the outer world (see, for instance, the different degrees of
opening shown by the reference objects of use-type 2 from open containers
to tweezers).
Linguistically relevant interpretation differences come about by the
operation of principles like those elaborated above, i.e., the mental
manipulation of given regions, their projection onto another kind of
214
Structure of the category of regions dans can select and

the cognitive principles underlying the subsumption processes to be assumed
( ) ^ ^ e s u b s u m p t l o n on the basis of processes of "Gestaltschlieung"

( 2 ) o subsumption on the basis of mentally created boundaries with unbounded entitles
(3) s u b s u m p t i o n on the basis of treating materially occupied regtons and their
outer surfaces on a par with hollow regions which are bounded materially
use tvDe 5:
selected region:
- 2-dlmenslonal
- bounded
(4)
la boule dans le cercle
use type S.
selected region:
- one-dimensional
I'tntervalle
dans la llgne
use type 6:
selected region:
- 3-dtmenslonal, but only two
dimensions relevant
- bounded
les maisons
dans cette
region
(4)
use type 7;
selected region:
- 2-dlmenslonal or only 2 dimensions relevant
- vaguely bounded, partially bounded or unbounded
les maisons dans le volslnage
(4)
(5) i " "
subsumptlon on the processes of mapping onto 2 or 1 dimenslon(s)

selection of the relevant dimensions ^ r e d u c t i o n of dimensions)
215
216
dimensionality or their metaphorical reconstructuring and similar principles.

My own theoretical interest in the analysis given above was three-fold:
1) I wanted to examine the role object conceptualization and categorization play in prepositional semantics. As it turned out, certain principles
operating in that domain play a crucial part in producing the interpretation differences of spatial prepositions.
This, on a more general cognitive linguistic level, was meant as a contribution to the examination of the interaction between the semantic and
the conceptual level.
2) From a more semantic point of view it was the aim of this paper to
show, by analysing dans as an example, that certain kinds of "polysemies"
in the realm of relational expressions can be quite naturally explained by
taking into account the operation of the kind of general cognitive principles elaborated above.
3) Methodologically, the type of analysis here propsed could serve to
check the ontologies assumed in recent semantic models. For instance,
in the semantics of spatial expressions semantic representations are often
defined in terms of concepts that are mostly either intuitively determined
or are mathematical or physical concepts. The analysis proposed in this
paper could check whether these concepts are really those relevant in the
production and interpretation of these lexical units.
Notes
1. See also:
les gens dans la rue
'people in the street'
vs.
un petit objet sur la rue
'a little object on the road'
qqn. dans I'escalier
'someone in the staircase'
vs.
qqc. sur I'escalier
'something on the stairs'
The same phenomenon can also be found with other prepositions, as for instance with
dans and sous as well as with dans and entre or parmi and others.
217
II est dans I'eau

'He is in the water'
vs.
II est sous I'eau
'He is under water'
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14,
15.
16.
17.
une tabatiere dans ses doigts

'a tobacco-box held by his fingers'
vs.
une tabatiere entre ses doigts
'a tobacco-box between his fingers'
See the basic assumptions in M. Bierwisch . Lang (1987: 2-3).
See . Bierwisch - . Lang (1987: 6-7 and . Lang 1989: 5-6).
R. Jackendoff (1983: 19).
See D. Wunderlich (1982: 10).
M. Herweg (1988: 69).
More precisely, it is the concept of the reference object we have to do with here, because
DANS* is processed with the interpretation of the noun phrase which is the inner
argument of dans and the contextual knowledge influencing this interpetation. The
result is an object concept (on the conceptual level).
As this paper deals only with the different interpretations of spatial dans, we have left
out of consideration that with some of the examples other prepositions such as, for
instance, entre, sur, , or en can be used or are even more current.
See again Note 8. Not all informants accepted this example, but others did. Even if it
is marginal it seems possible and interpretable.
This is a term introduced, as far as I know, by A. Herskovits (1985) in her prototype
model of the semantics of English spatial prepositions. As will be seen below, the term
is not used here in quite the same sense in which it is used in A. Herskovits or in B. W.
Hawkins (1986). Use-types here are just classes of interpretations of dans, the classification criteria being the different properties of the dans-region identified. Use-types are
different interpretations (and belong to the conceptual level), not different meanings
in the sense of different lexical entries for dans.
What I present in this paper is the result of the analysis of a corpus of more than two
thousand constructions with dans collected from modern French oral and written texts
and from linguistic literature on prepositions. The examples have all been controlled
by French informants and they have been standardized by reducing them to the form
(NP Prep NP). But the results of this analysis can easily be extended to more complex
constructions because the contextual influence of verbs and other elements of the
context seem to operate on the interpretation of (Prep NP), as will be shown below.
Relations like the one between a text or a picture printed in a book, and the book or
beween something reflected in a mirror, and the mirror, and similar relations are left
out of consideration in this paper, because they are more complex relations than the
simple localization relations analysed here.
See again Note 12.
See again Note 9.
The principle operating in these cases could be added to the principles elaborated in
this paper. But it is a principle of an even more general kind that does not only operate
on concrete spatial entities.
See again Note 8.
See M. Herweg (1988: 74).
218
References
Ballmer, Thomas. . Roland Posner (eds.)
1985
Nach Chomskysche Linguistik. Neuere Arbeiten von Berliner Linguisten.
Berlin, New York: W. de Gruyter.
Bayer, Josef (ed.)
1988
Grammatik und Kognition. Psycholinguistische Untersuchungen. Opladen:
Westdeutscher Verlag.
Bierwisch, Manfred
1986
"On the nature of semantic form in natural language", in: F. Klix H.
Hagendorf (eds.), 765-784.
1987
Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven. Berlin:
Akademie-Verlag.
Habel, Christopher Michael Herweg Klaus Rehkmper (eds.)
1989
Raumkonzepte in Verstehensprozessen. Interdisziplinre Beitrge zu Sprache
und Raum. Tbingen: Niemeyer.
Hawkins, Bruce W.
1984
The semantics of English spatial prepositions. San Diego, Cal. [Ph.D. Dissertation, University of California.]
Herskovits, Anette
1986
Language and spatial cognition. Cambridge: Cambridge University Press.
Herweg, Michael
1988
Zur Semantik einiger lokaler Prpositionen des Deutschen. berlegungen zur
Theorie der lexikalischen Semantik am Beispiel von in, an, bei und auf. LILOGReport 21. Stuttgart: IBM.
1986
Die Semantik lokaler Prpositionen. Ein prototypensemantisches Modell fr
die franzsische Prposition dans mit einer Analyse der Beziehungen zwischen
der Prposition und den Objektbezeichnungen in den Prpositionalsyntagmen. [Habilitationsschrift, Universitt Konstanz.]
Jackendoff, Ray
1983
Semantics and cognition. Cambridge, Mass.: MIT-Press.
Klix, Friedhart Herbert Hagendorf (eds.)
1986
Human memory and cognitive capabilities. Mechanisms and performances.
Amsterdam: North-Holland.
Lang, Ewald
1987
"Semantik der Dimensionsauszeichnung rumlicher Objekte", in: M. Bierwisch - E. Lang (eds.), 287^158.
1989
"Primrer Orientierungsraum und inhrentes Proportionsschema", in: Christopher Habel Michael Herweg Klaus Rehkmper (eds.), 150-173.
1988
"Gestalt und Lage rumlicher Objekte: Semantische Struktur und kontextuelle Interpretation von Dimensionsadjektiven", in: J. Bayer (ed.), 163-191.
Schwarze, Ch.
1982
"Stereotyp und lexikalische Bedeutung", Studium Linguistik 13: 1-16.
1984
"Observations on lexical contrasts and text understanding", in: G. Ldi
. Strieker I. Wuest (eds.) "Romania ingeniosa", Festschrift Professor Dr.
D. G. Hilty zum 60. Geburtstag.
219
Vandeloise, Claude
1986
L'Espace en franfais. Semantique des pripositions spatiales. Paris: Editions du
Seuil.
Wunderlich, Dieter
1982
"Sprache und Raum", Studium Linguistik 12: 1-19 and 13: 37-59.
1985
"Raumkonzepte Zur Semantik der lokalen Prpositionen", in: . T. Ballmer
- R. Posner (eds.), 341-351.
Semantic and conceptual aspects of the preposition

durch1
Ingrid
Kaufmann
1. Introduction
In this paper I will present an analysis of the German path preposition
durch 'through'. This analysis is be understood as an exemplification of
the general ideas about local prepositions presented in Wunderlich (1991).
I will just give a short overview of the main ideas; for details I refer to
Wunderlich's paper.
As a general framework we assume a two-level approach to semantics,
with semantic form (SF) on the one hand and conceptual structure (CS)
on the other. While SF is part of the grammar, and therefore interacts
with syntax and morphology, CF is extralinguistic. Categories and structures in C F are based on human experience in dealing with the world as
well as biologically motivated phenomenona such as perception and
motion.
The interpretation of an expression results from the interaction of both
levels. The semantic form of a lexical item is assumed to consist only of
the information that plays a role in every specific use of that item. The
variation in interpretation is due to conceptual information associated
with the item, which is activated by the relevant context. To allow for
this, semantic structure contains parameters that have to be fixed by
conceptual information with respect to a relevant context.
1.1. General assumptions about the semantic form of spatial prepositions

We assume that there is one component shared by all spatial prepositions.
This component carries the information that an object is located in a
region that has to be defined with respect to an object y. In (1) LOC is
the location relation between an individual and a region. R E G (y) is a
metavariable that stands for a specific neighbourhood region of the
individual y.
222
(1)
Ingrid Kaufmann
Xy (LOC (, REG (y)))
Some of the static prepositions, such as in and under, can be dealt with
by just specifying the function variable REG. Others such as on and at
carry more information: they have an additional component constraining
further the relevant relation by requiring contact between and y. We
assume that the semantic form of spatial prepositions can be generally
represented by a semantically decomposed structure consisting of a location component and another component which imposes further constraints.
1.2. The ontological character of "path"

Before I go into a detailed analysis of durch, I would like to explain our
view on the notion of "path". It can be shown that there is no property
which can be used to characterize the notion of path in general. Instead
there are two path properties which can be constructed by an ordering
and a connection relation on regions, and at least one of them has to be
met. To support this claim, I will briefly examine the path properties that
are generally assumed to be important: linearity, continuity and direction.
In fact, for each one of these properties examples can be found in which
it is not fulfilled.
(2)
a. Von den Bergen strmte das Schmelzwasser in das Tal.

The water flowed into the valley from the mountains.
b. Die Wolkendecke breitete sich langsam ber den Himmel aus.
The cloud spread slowly over the sky.
c. Nach allen Seiten zogen Menschen aus der Stadt.
People left from every side of the city.
The examples show that linearity is not a property that a preposition

requires its path to have, but just an effect of the gestalt properties of
the regions defining the source and the goal of the path, and the gestalt
properties of the moving object.
Continuity:
(3)
a. Er stellt das Glas auf den Tisch.

He stands (puts) the glass onto the table,
b. *Er stellt den Stuhl durch den Flur.
He stands (puts) the chair through the corridor.
Semantic and conceptual aspects of durch
223
Causative positional verbs such as stellen, setzen, legen 'put' do not deliver
a continuous path but only a change between two states of location. Not
all of the path prepositions can be combined with them, as example (3 b)
shows. However prepositions such as auf 'on'; in 'in'; an 'at' and vor 'in
front o f , which carry the morphological feature [ + Dir] and therefore
govern accusative case, can be combined with this class of verbs (3 a).
For these prepositions (P [ + Dir]) then, a continuous path is not necessary, whereas for durch it is.
Direction:
(4)
a. die Mauer durch Berlin

the wall through Berlin
b. der Zaun ber den Berg
the fence over (acc) the mountain
Durch and ber allow for an external argument that provides an undirected path. In the case of ber the situation is even more complicated.
The example in (2 b) shows that ber allows for a non-linear path, while
(4 b) shows that it allows for a non-directed path as well. The examples
in (5 a) and (b) show that it does not allow for a path which is neither
linear nor directed. This is not surprising, since this kind of spatial
configuration is expressed by the static prepositions ber or auf.
(5)
a. *die Wiese ber den Berg

the meadow over (acc) the mountain
b. * die Wolken ber den Berg
the clouds over (acc) the mountain
c. die Wiese auf dem Berg
the meadow on (dat) the mountain
d. die Wolken ber dem Berg
the clouds over (dat) the mountain
One can conclude that there are different properties which can constitute
a path, but that none of them is really basic. What is basic is only the
condition that more than one region must be involved. Now there are
two possible relations between these two regions: an ordering relation
and a connecting relation. Using these, the relevant path properties can
be built up. By relating the two regions with the ordering relation, one
gets a non-continuous, directed path. By relating them with the connecting
relation, one gets an non-directed, continuous path. By applying both,
one gets a directed, continuous path. Linearity results if an object which
224
Ingrid Kaufmann
is conceptualized as being zero-dimensional (punctual) constitutes the

path or if two punctual regions are connected.
Under these assumptions the conceptually relevant notion of path
reduces to the very simple concept of a related regions. This concept can
be motivated by the human perception of the world: regions can be
grouped together to build up complex regions of a specific shape, or they
can be viewed under a specific perspective that creates a direction.
In other words, ontologically there is no such thing as a single object
"path", so when I talk about prepositions requiring a path or nouns and
verbs providing paths, "path" is to be interpreted as some instantiation
of the relevant properties. An object can instantiate a path if it shows up
one of the properties mentioned above. This does not mean however that
it is able to provide a path in every context, but only in case exactly this
property is needed. Though I tried to show above that the conceptually
relevant path properties are connectedness and order, that is, certain
relations between regions, I will stick to the common notions of continuity
and directedness when talking about objects functioning as paths and
about prepositions imposing constraints on their arguments.
1.3. Consequences: The parameter D (x)

Rejecting the existence of a unique concept of path has consequences for
a semantic analysis of path prepositions. If there is no ontological object
defined as path, it is impossible to deal with path prepositions by postulating a semantic variable that denotes a path, or by assuming that
complex PPs denote paths. On the other hand, path prepositions do
constrain their external argument in a certain way and this must be
accounted for in semantic form. In a two-level approach to meaning this
can be done by means of parameters. Parameters are constants in semantic
form that, depending on the context in question, have to be fixed by
conceptual information. The conceptual units specifying the parameter
can differ to a certain extent but have to meet certain conditions that are
regulated by specific mapping constraints.
To account for the fact that the properties of a path vary with respect
to the lexical item that provides it and that prepositions differ in the
requirements they impose on the path, we assume a path parameter D
(x) in the semantic form of all path prepositions. D (x) is to be read as
"a dimension to be found relative to (the external argument of the
preposition) x". This parameter ensures that the external argument of
225
the preposition cannot be saturated by just any noun, but only by those
nouns that provide the information required. If the conceptual information associated with a noun cannot provide a path, the parameter
cannot be instantiated and the expression remains unacceptable. The
advantage of taking a parameter D (x) is obvious with attributive constructions.
(6)
a. der Weg durch den Wald

the path through the wood
b. die Reise durch den Harz
the journey through the Harz
c. die Leitung durch den Garten
the piping through the garden
d. der Bus durch Kln
the bus through Cologne
e. der Spie durch den Braten
the spit through the roast
The examples in (6) show that the external argument of durch can be
saturated by nouns denoting quite different objects. These objects,
though, have one important feature in common: they are all associated
with a path. However, the specific properties of this path and the role it
plays with respect to the noun vary considerably. The nouns in (6 a), (c)
and (e) all provide a path-like gestalt. Apart from that, though, (a) and
(c) also contain the information that there is some kind of object moving
along this path, whereas in (e) there is no such information at all. The
path does not necessarily have to be provided by the gestalt of the object,
as can be seen from the example in (d). One cannot conceive of a bus as
having a gestalt that could in any way provide a path. Moreover, the
gestalt property of an object is obviously not sufficient to determine
acceptability, as can be seen from the examples in (7):
(7)
a. ?der Teppich durch den Flur

the carpet through the hall
b. ?der Stock durch das Zimmer
the stick through the room
The examples in (8) may serve as illustration for the great variety of
nouns that can provide a path.
(8)
a. die Tr in den Garten

the door in(to) the garden
226
Ingrid Kaufmann
b. das Visum in die USA

the visa in(to) the USA
c. die Fhre nach Helgoland
the ferry to Helgoland
d. der Blick durch den Flur
the look through the hall
e. das Fenster zum Hof
the window (on)to the yard
f. der Schrei aus dem Wald
the cry from the wood
g. die Mauer durch Berlin
the wall through Berlin
In (8 a)-(8 c) the external argument is associated with a concept of motion
but plays different roles within this concept. The relevant path is part of
the concept of motion, too. (d) and (e) involve the concept of perception.
Here the Observer Axis, as assumed by Lang (1987), can be regarded as
the relevant path. While this Axis is directly associated with look (as its
gestalt property in (8 e)), in the case of window it can only be associated
by means of functional information, (f) involves a very abstract path that
must be reconstructed by a source region, defined by the person crying,
and a goal region, defined by the listener. The path in (g) does not entail
direction but is only defined with respect to the gestalt of the located
object.
The data show that it is impossible to deal with the external argument
of the preposition by restricting it to a specific semantic type "path". The
variety of combinations supports the assumption that there must be
access to conceptual information which provides this path. Assuming a
parameter D(x) allows one to deal with these examples. The external
argument itself does not have to denote a path, it only has to provide
one by means of the conceptual information it is associated with.
2.
durch: the analysis
2.1. The location component

First it has to be shown how the assumption that a path preposition like
durch contains a location component can be motivated. Obviously there
is a close connection between a situation such as that described by (9 a)
and the one described by (9 b).
(9)
227
a. Er wandert durch den Wald.

He wanders through the wood,
b. Er wandert im Wald.
He wanders in (dat.) the wood.
The sentence in (a) implies that for some part of the situation in question
the description in (b) is true as well. This relation between durch and in
holds for every use of durch. It is, however, not symmetric: a situation
of wandering inside a wood does not contain a situation of wandering
through a wood. This shows that, as well as containing the same information as in, durch also imposes further conditions not required by in.
This relation between in and durch can be accounted for by incorporating
the semantic form of in in the representation of durch. There are, though,
alternative representations that can deal with this relation as well. For
example, a representation of durch in which only the path is located in
the interior of an object can also account for this implication, since the
moving object is situated on the path and therefore has to be included
in the same region as the path.
In fact, there is another observation which shows that, as in the case
of in, the located object itself must be the argument of the locating
relation LOC.
(10)
a. Er geht durch die Wiesej??den Rasen.

He walks through the meadow/the lawn.
b. der Weg durch die Wiese/den Rasen
the path through the meadow/the lawn
c. Er geht durch den Kreis/den Marktplatz.
He walks through the circle/the market square.
(11)
a. Er steht in der Wiesej??dem Rasen.

He stands in the meadow/the lawn.
b. der Klee in der Wiese/dem Rasen
the clover in the meadow/the lawn
c. Er steht im Kreisj??im Marktplatz.
He stands in the circle/the market square.
A look at the examples in (10) and (11) shows that durch, like in, imposes
certain conditions concerning the dimensionality of the reference object.
The examples suggest that the reference object must provide at least those
dimensions that are provided by the located object in order to include it.
That is, a two-dimensional object such as a lawn cannot include a threedimensional one. To account for these dimensionality conditions the
228
Ingrid Kaufmann
located object itself must be accessible. If only the path on which the
object is moving is included in the relevant region, one cannot refer to
the dimensions of the located object.
However, there are exceptions involving geometrical terms as in (10 c)
and (11c). These exceptions can be dealt with by assuming that the
relevant dimensionality properties are determined by the internal argument of the preposition. Geometrical objects such as circles and squares
are defined only in two dimensional space. Owing to this, inclusion can
refer only to the edge of the circle. "Real objects", even if they just have
two dimensions (such as surfaces and market squares), are defined with
respect to three-dimensional space. In contrast to purely geometrical
entities two-dimensional objects are subject to different conditions, determined by the real world they are part of.
In what follows, I take it for granted that the identical dimensionality
conditions imposed by in and durch justify the incorporation of the
semantic structure of in into the semantic structure of durch. One part of
the semantic structure of the lexicon entry of durch is therefore the
location component as represented in (12).
(12)
y (LOC (x, INT (y)))
INT (y) must be interpreted as the interior of y, that is, the relevant parts
of are surrounded only by the inner points of y. The weakening from
inclusion of the whole object to inclusion of the relevant parts only is
necessary in order to deal with examples such as die Pfeife im Mund in
the case of in and der Spie durch den Braten in the case of durch, which
show that it is not the whole object that has to be included in the relevant
region but only those parts of the object that are focussed upon. (Compare
Herweg 1988 for a discussion of this kind of variation in interpretation.)
2.2. The path component

There are a number of problems connected with the representation of
the path component of durch:
1. The conditions imposed by durch on the path-properties must result
from this component.
2. The effect of a durch-PP on the structure of the situation must be
accounted for.
The first point involves various aspects that have to do with the compositional properties of durch. Unacceptable constructions result if the
229
requirements of durch are not met. Therefore the combinatory properties

of durch can serve as a test for determining semantic structure. To work
out the specific requirements of durch I will contrast it with the [ + Dir]
prepositions {in, an, auf etc. [acc]), which impose different constraints on
the path they require. The second point primarily concerns the effect of
durch-PPs functioning as arguments of verbs of motion, since the temporal
structure of the complex situation always reflects the spatial structure
durch.
Before I turn to the semantic representation of the path component of
durch, I would like to discuss two points: (i) what part of the information
present within the external argument provides the required path and (ii)
how does this information specify the parameter D (x). In the course of
this discussion the specific requirements imposed by durch will show up.
Since these constraints result from the semantic representation of the
preposition, they must be taken into account when a proposal for semantic structure is made.
3. The interpretation of D (x)

3.1. The domain of specification of D (x)
The conceptual information required to instantiate D (x) may be provided
by different elements, depending on the particular phrase concerned. In
constructions in which the PP is the argument of a verb, for example,
this information is provided by the verb itself but in attributive constructions it comes from the noun. (13) shows examples of these two different
types of constructions, in both of which however the located object is
the same. A comparison of the two constructions shows that the required
information does not come from just any element within the sentence:
(13)
a. die Leitung durch

the piping/wiring
b. ?die Leitung liegt
the piping/wiring
den Garten
through the garden
durch den Garten
lies through the garden
In (13 a) the value of D (x) can be specified by the noun; this is not
possible in (b). Obviously some sort of constraint must be imposed on
the domain of specification. This constraint can be formulated as follows:
230
(14)
Ingrid Kaufmann
The domain of specification

The first maximal projection that dominates the PP constitutes
the structural domain for instantiation of D (x). The head of
this construction instantiates the parameter.
Note that specification by the head of the dominating maximal projection

reflects semantic composition. The semantic structure of the first maximal
projection is, in fact, the first structure accessible from the next step in
composition.
Under common assumptions about syntactic structure, (14) leads to
the correct result in every case: in attributive constructions specification
is by the noun; in constructions in which the PP is an argument of the
verb, specification is by the verb, and (13 b) is blocked. This should not
exclude the possibility of having D (x), in certain contexts, specified by
some other information. (Compare the sentence in (15), which was uttered
in a discussion about the problems of travelling to the GDR).
(15)
Hunde in der DDR brauchten immer einen Impfpass

Dogs in(to) the GDR always needed a vaccination certificate
book
Such utterances however cannot be interpreted on the basis of the structure involved, that is, according to the principles of semantic compositionality. In fact, if uttered in isolation, such constructions are generally
rejected.
As mentioned above, specification of a parameter is carried out by
conceptual information provided by the context. In the following, I will
discuss in more detail how specification of D (x) by different verbs and
nouns can take place.
3.2.
Specification of D (x) by the verb
3.2.1. Verbs of motion

For verbs of motion I assume a semantic representation containing two
central components: (i) the predicate MOVE (x) and (ii) a predicate M O D
(x), which encodes the specific kind of motion.
The verb gehen 'walk' is represented roughly as in (16). Following
Bierwisch, I assume that these verbs subcategorize for a predicate. is
the predicate variable to be instantiated by the semantic representation
delivered by the local prepositional phrase.
(16)
231
Xs (MOVE () & GEH () & ()) (s)
In the case of verbs of motion the component that is relevant for

instantiating D (x) is the one which encodes movement, that is, the
predicate MOVE. MOVE must be conceptually interpreted as a function
that assigns local regions to an individual at different periods of time.
That is, mediated by the moving object, there is a mapping from the time
axis into space; thus a path is constructed which has the same structure
as the time axis and is therefore directed. The resulting path satisfies the
conditions required for the instantiation of D (x).
3.2.2. Causative positional verbs

Causative positional verbs do not allow combination with durch. Unlike
verbs of motion, causative positional verbs do not contain the predicate
MOVE:
(17)
Xy (CAUSE (, (BECOME (MODy)) & (y))) (s)
The BECOME component gives rise to a two-phase situation consisting

of an unspecified state changing into a specified resulting state. The initial
state is undefined, in the sense that it can only be determined relative to
the resulting state, in fact as its negation. The obligatory theta role
can only be saturated by a PP that specifies the resulting state of the
located object, since the location of y in a heterogeneous situation containing a change in position cannot be static. A two-phase location event
results when the verb combines with the PP: there is a state at which y
is not yet located within the relevant region and a state at which location
has already taken place. These local states are sufficient to fix the points
which define the path. Directionality, once again, results from the temporal structure. Being part of one and the same event, the points are
connected indirectly and no continuous spatial path is involved. This is
sufficient to fix the parameter of the [+ Dir], such as in, an, auf [acc],
since in this case only the initial and resulting local states are relevant.
However, this is not sufficiant to fix the path required by durch: when
durch is used with causative positional verbs, as in (18 a), reinterpretation
takes place and the only possible interpretation is the resultative one.
One never gets the interpretation that a path is included in the object of
reference, since such an interpretation, which is always primary when
MOVE is present, requires a continuous path.
232
(18)
Ingrid Kaufmann
a. Er stellt den Stuhl durch das Fenster.

He stands (puts) the chair through the window,
b. * Er stellt den Stuhl durch den Flur.
He stands (puts) the chair through the hall.
3.2.3. Positional verbs

The location carried out by positional verbs is static and therefore they
do not have a component that delivers a path. However they do provide
specific information about the position of the object located in space.
This information must be used to explain why verbs like hngen 'hang'
and ragen 'protrude' can be combined with directional PPs, but not sitzen
'sit'; stehen 'stand'; liegen 'lie' and knien 'kneel'.
(19)
a. Der Vorhang hngt durch das Fenster.

The curtain hangs through the window.
b. ?Der Teppich liegt durch die Tr.
The carpet lies through the door.
c. Die Leute stehen durch den Park.
The people stand through the park.
I assume that a directional PP does not specify the same information as

a static PP: The location component of a positional verb relates only to
the region of attachment between the located object and the reference
object. This component is specified by static local information. On the
other hand, under certain conditions the position component of the
positional verb can be specified by directional PPs. In the case of ragen
and hngen this component delievers direction, in the case of liegen,
stehen, etc. it does not. This is why hngen and ragen allow for a
directional PP while the others do not.
(20)
a. Die Lampe hngt an der Decke.

The lamp hangs at the ceiling,
b. Die Lampe hngt von der Decke.
The lamp hangs from the ceiling.
(21)
a. Der Vorhang hngt auf den Tisch.

The curtain hangs on the table,
b. ??Das Bild hngt auf den Tisch.
The picture hangs on the table.
(22)
a. Der Ast ragt in das Zimmer.

The branch protrudes into the room.
233
b. *Der Ast ragt am Baum.

The branch protrudes at the tree.
(20) illustrates the two possibilites in specifying the positional verb hngen.
(21) shows that if the positional component is specified, the located object
has to provide certain gestalt properties. This is never the case if only
specification of the location component takes place. The examples in (22)
should illustrate that it is really the specification of the positional component that is carried out by the directional PP: ragen lacks a location
component and it cannot be combined with a static PP. That is, the
directional preposition can only specify the positional component.
On the basis of these assumptions one can explain why the combination
of (nondirectional) positional verbs with PPs [ + Dir] is not possible. The
positional component lacks direction and can thus not specify the required
path. However the example in (8g) showed that direction is not always
required by durch. Why the combination between positional verbs and
durch is still blocked will be explained later.
3.3. Specification of D (x) from information within the noun

With respect to the information carried by the noun I assume the twolevel framework proposed by Bierwisch (1983), in which the semantic
representation encodes the information common to all possible readings
of a lexical item. This basic representation can be further differentiated
by means of the various concepts associated with the noun.
With respect to the parameter D (x), the factors relevant for nouns
denoting concrete objects are gestalt properties as well as properties that
are associated with the function of the object ("functional properties").
The term "function" should not be interpreted too strictly in this context.
While the Object Schemata developed by Lang (1987) provide a framework which allows one to deal with gestalt properties, there are hardly
any concrete ideas about the representation of other kinds of information
present within the noun. One way of dealing with the functional information of the noun is to assume stereotyped concepts which are associated
with a noun. A concept of motion, for example, could be such a stereotype, involving a source, a goal, a path, a moving object (theme), an
instrument etc. In such a stereotyped concept, the role of the object
denoted by the noun can be fixed. The representations suggested below
rely mostly on an intuitive analysis and are not to be taken too seriously
234
Ingrid Kaufmann
until further work is done about the cognitively relevant categories of

information concerning the function of objects.
In the following I deal with different classes of nouns which vary with
respect to the nature of the information for instantiating D (x).
3.3.1.
Specification by gestalt properties
3.3.1.1. Directed objects

(23)
a. die Strae durch den Park

the road through the park
b. die Leitung durch den Garten
the piping through the garden
In those cases in which D (x) is specified by the noun, the relevant

information can only come from the components which have spatial
properties: the spatial gestalt properties of the denoted object, any functional properties which involve some sort of spatial concept, and structural properties. Gestalt properties are encoded in the Object Schema
associated with the noun, which entails information about the object's
axes. The maximal axis is the only object axis that satisfies the requirements of D (x): because of the prominence of this axis, the whole object
can be viewed as a projection onto one axis. Although the presence of a
maximal axis is a possible condition for the specification of D (x) by
gestalt properties, it is not sufficient. Some sort of functional information
concerning space must also be associated with this axis. Bleistift 'pencil'
is an example of an object that has a maximal axis and provides a
functionally fixed "directedness', relative to this axis, which cannot however be interpreted spatially. Due to the functional information "instrument for writing", there is an asymmetry between the two ends of the
pencil that cannot be described in terms of a spatial ordering. As a result,
the spatial property of directedness required for the interpretation of the
phrase der Bleistift auf das Papier 'the pencil onto the paper' cannot be
assigned, so that the phrase is meaningless.
The following constraint can be devised to rule out such phrases:
(24)
The information that is used to specify a parameter must

originate from only one conceptual domain. In the case of D
(x) it has to involve spatial properties.
The function of the two modified nouns in (23) can be described as

"medium of motion". A track, for example, is used by people to walk
235
on and a pipe allows water to pass through. One can therefore assume
that the functional information of these nouns entails the notion of
motion in space, which can be accessed to specify D (x). The path defined
by the concept of motion runs parallel to, on, or in the maximal extension
of the object and is therefore associated with it. In the case of nouns
whose spatial function refers to the maximal extension, specification of
D (x) can be carried out by the maximal axis.
The relevant concepts connected with track and piping can be represented as in (25) below. In the first line some of the gestalt properties are
given: "max" refers to the maximal axis of an object, "dist" to "an object
axis perceived as inside diameter of a hollow body" (Lang 1990: 135),
"vert" refers to the vertical, "sub" to that axis of an object which is
perceived as its thickness, and 0 refers to a disintegrated axis which is
not further specified. The notation used below is a rather simplified form
of the representation developed by Lang, but it is sufficient for the
purpose of illustration. (Cf. Lang for the structure and content of Object
Schemas (OS).) In the second line a kind of "concept of motion" is
suggested. It is represented here as a predicate with a list of arguments
which can be interpreted as thematic roles, "t" refers to the moving
object, that is to the "theme", and "path" to the route taken by the
moving object. In some contexts it might be useful to refer to the goal,
source, instrument of motion etc. I leave open the question of which
roles are involved in a concept of motion. This kind of representation
does not reflect any particular theoretical assumptions, it could be a
propositional representation as well. The items in bold type mark the
role that is played by the specified noun in the functional schema. There
may be further associations between the different levels, such as the one
between "path" and "max", which means that the maximal axis of the
object is the one which serves as a path, that is, movement takes place
along max.
(25)
a. track:
gestalt properties:
OS: (max, 0)
functional properties: MOVE (th, path, ...)

b. piping.
gestalt properties:
functional properties:
OS (max, dist)
MOVE (th,
I path, ...)
(liquids, gas, ...)
236
Ingrid Kaufmann
(25) shows how different levels of information associated with one lexical
item can be connected in conceptual structure. Only those parts of the
conceptual schema that are associated with each other can interact in
specifying a parameter.
3.3.1.2. "Bounding objects"
(26)
a. der
the
b. die
the
Zaun durch den Wald

fence through the wood
Mauer durch Berlin
wall through Berlin
With nouns that, because of their lack of direction, can only be combined
with non-directed path prepositions such as durch, um 'around' und lngs
'along', another kind of functional information is involved. I would like
to call the relevant function " B O U N D A R Y " (or INTERSECTION for
durch), and the objects associated with this function "bounding objects".
As in the case of the directed objects above, the functional property of
nouns like Zaun and Mauer (fence, wall) is associated with gestalt properties. For example, with Zaun the spatial bounding of an object takes
place relative to the maximal axis of the fence, i.e., its length, and not
in relation to its height or width. Moreover, BOUNDARY/INTERSECTION requires that the bounding/intersecting object should have a
certain minimal length relative to the size of the bounded/intersected
object y. Ideally the maximal axis of should be identified with one of
the sides (in the case of durch with an inner segment) of y. As before D
(x) is specified by the maximal axis of x.
(27)
fence, wall.
gestalt properties:
OS: (max, vert, sub)
functional properties: BOUNDARY (bounding obj., region,...)

Note that the length of the intersecting object relative to the intersected
object is a relevant factor in the use of durch. Only a fence that provides
this length can be described by (26 a), but not a 20 m long fence inside a
wood 50 m in diameter. This reveals an important property of durch: The
path must reach the borders of the region in which the located object is
included, that is, the object must be completely traversed by the path. I
will call this requirement of durch "condition of complete traversal". Note
that this is not an independent constraint as, for example, the constraint
in (25), which regulates access to conceptual information with respect to
237
the specification of D (x). It is only a condition that appears as an effect

of the semantic form of durch. Whatever the semantic form of durch
looks like, the condition of complete traversal must be derivable from it.
To summarize, one central aspect of the semantics of durch is that
complete traversal takes place. Thus, the located object has to provide a
path that is able to fulfill this condition. Now there are two alternatives:
The first is that the path is directed, that is, it is supposed to have a goal;
in this case complete traversal can be assumed because the region traversed defines just a part of the path leading to the goal. In the second
case the functional information of the noun guarantees that the located
object intersects the reference object, in which case direction is not
required. Note that the [+ Dir] prepositions (as in, an, a/[acc]) do not
allow for this second possibility: due to their semantic representation they
require a directed path to instantiate D (x).
Now this condition of complete traversal can be used to rule out
examples such as those in (28), as well as the combination of non-directed
positional verbs with durch. Carpets and rails are neither bounding
objects, nor are they directional; thus they cannot guarantee complete
traversal. Similarly positional verbs only provide information about gestalt and positional properties, as do objects such as carpets and rails.
Therefore the combinations are blocked for the same reason.
However, as the examples in (29) show, if an element such as quer
'crosswise', 'diagonally' is added, the resulting constructions are completely acceptable since quer itself delivers the information that complete
traversal takes place.
(28)
a. ?der Teppich durch den Flur

the carpet through the hall
b. ?die Stange durch das Zimmer
the stick through the room
(29)
a. Die Stange quer durch das Zimmer (wurde fr Ballettbungen

angebracht)
the bar diagonally through the room (was installed for ballet
exercises)
b. Er liegt immer quer durch das Bett.
"He lies always diagonally through the bed."
c. Die Schlange steht mitten durch den Supermarkt.
"The queue stands middle through the supermarket."
238
Ingrid Kaufmann
3.3.2. Specification by the concept of a functional path

(30)
a. der Bus durch Kln

the bus through Cologne
b. ??das Auto durch Kln
the car through Cologne
For the nouns in these examples, like those in (23), the relevant information for the instantiaton of D (x) is associated with the concept of
motion coming from the functional information. However, here the path
is not associated with the gestalt properties of the located object since it
functions as a means of conveyance and therefore has to be interpreted
as the theme of motion. The attributive constructions in (30) suggest that,
besides the path provided by the concept of motion, additional criteria
are required since Auto and Bus are both instruments of conveyance, and
therefore have access to the same concept of motion. The relevant difference seems to arise from the fact that a bus generally has a fixed route,
while a car is not constrained at all in this respect. The additional criterium
is therefore route boundedness.
Up to now, the only relevant factor for movable objects to fulfill the
condition of complete traversal has been directedness. Therefore the
question now arises as to why route boundedness is required in attributive
constructions. I do not want to discuss this point in detail, but only to
give a hint at a possible explanation. Attributive modifiers such as those
in (30) have no access to temporal information. Due to this there are two
possible interpretations: the attribute characterizes the state of the object
at the moment of speaking or it gives a general characterization of the
object. Now, for movable objects it is not possible to move along a path
in one moment, so the description in (30) can never be used to describe
the momentary state. As a result, these constructions have to be interpreted as characterizing a permanent property. Movable objects that are
not bound to specific routes cannot be appropriately characterized by a
PP referring to a path. Thus, the constraint of route boundedness on
movable objects is just an effect of attributive constructions.
(31)
a. car.
gestalt properties: OS: <max, 0, vert)
(human)
I
functional properties: TRANSPORT (instr, th ...)
I
MOVE (th, path, ...)
239
b. bus:
gestalt properties: OS: <max, 0, vert)
(public)
(human)
I
I
functional properties: TRANSPORT (instr, th, ...)
I
MOVE (th, path, ...)
I
(fixed route)
The representations in (31) show that both cars and buses are instruments
of transport for people and are therefore moving objects. While in the
case of car there is no further information about the path, in the representation of bus the path is restricted to one specific route.
3.4. The conditions on the specification of D (x): a summary

The specification of D (x) within a suitable domain is constrained by
conditions on the path imposed by the specific preposition. In the case
of durch this condition is the requirement for a path to traverse the
relevant region completely. This condition has different effects on the
properties of the path specifying D (x), depending on the character of
the external argument of the preposition. If complete traversal is guaranteed by functional information (intersection), no further requirements
must be met. If not, the path must be directed in order to fulfill the
condition. In both cases D (x) can only be specified by a continuous
path. A general constraint on attributive constructions rules out the
modification of nouns denoting movable objects that cannot be assigned
a fixed route.
4. The various interpetations of durch:

problems concerning the internal argument
How can the parameter D (x) be integrated into the path component of
the semantic representation of durch? In discussing the nouns denoting
bounding objects, I said that the path specifying D (x) must be related
240
Ingrid Kaufmann
to a line of intersection. To find out what kind of relation this could be,
a number of phenomena must be dealt with.
1. In addition to the standard interpretations that have been analysed
up to now, there is another interpretation of durch that seems to contradict
the condition of complete traversal. This reading, which allows D (x) to
be fully included in the object of reference, is possible if the object of
reference is given by a bare mass or a plural noun. Moreover it can only
combine with verbs of motion. In contrast to the standard interpretation
of durch, which forces an accomplishment reading, the sentences in (32 a)
and (b) denote processes. This shows that in fact no complete traversal
takes place.
(32)
a. Er watet durch Matsch.

He wades through mud.
b. Er wandert durch Wiesen und Wlder.
He wanders through meadows and woods.
c. ??die Strae durch Matsch
the road through mud
The reason why in combination with masses complete traversal cannot

take place is simple: It does not make any sense to speak of an entity
leaving a mass, since masses do not have boundaries.
The fact that only verbs of motion can be combined with PPs whose
internal argument is a bare mass noun suggests that at least the possibility
of a complete traversal must remain open. Because of the temporal
unboundedness of process verbs, it is possible to have a continuous
approach towards the potential boundaries of the mass-object, which
however are not available in the present context. Note that this interpretation of masses is restricted to bare nouns. As soon as a definite
determiner turns up, the interpretation changes since the determiner
individuates a portion of mass with concrete borders.
2. Apart from these occurrences of durch there is another problematic
group of nouns that can appear as internal arguments of durch. These
nouns denote objects that have to be classified as two-dimensional. If
combined with durch, the path crosses these objects in the third dimension
(cf. (33)).
(33)
a. der
the
b. der
the
Blick durch das Fenster

look through the window
Schritt durch die Tr
step through the door
241
Including the standard interpretation, there are three spatial configurations that must be accounted for.
In the configurations in (a) and (b) the path is included in the object
of reference; in the configurations in (b) and (c) it extends beyond the
borders of the object of reference. Neither of these descriptions alone
covers all three configurations. Owing to this, it has often been assumed
that in semantic representation durch must involve the relation of improper inclusion of a part of the path inside the object of reference. Such
a representation does not seem to be appropriate since it cannot exclude
the expression in (32 c). Moreover, it is unsatisfactory because it does not
allow access to the object axes in the sense of Lang (1987). Obviously
there is a certain regularity as to which route the path takes, depending
on the gestalt properties of the reference object as encoded in the object
schema. Therefore an analysis which provides access to the object axes
seems preferable. The notion of intersection in fact does exactly this,
since intersecting lines can coincide with object axes.
While the configurations (b) and (c) in figure 1 fulfill the condition of
complete traversal, this does not hold for the mass interpretation. Since
mass nouns as internal arguments of PPs are restricted to combination
with verbs of motion, I assume that the other cases are basic, that is, the
condition of complete traversal holds. The interpretation with mass nouns
has to be explained by their specific properties.
In dealing with the mass interpretation, I assume that one has to
differentiate between the general unboundedness of masses and their
appearance in a certain context. That is, if we talk of mud in general, we
conceptualize an unbounded object, whereas if we talk about walking
through mud, we conceptualize an object which we know must have
boundaries somewhere. In the relevant context there are no boundaries,
but we know them to be somewhere beyond. In such a concept one can
242
Ingrid Kaufmann
imagine that the condition of complete traversal is fulfilled as long as the

path can extend until the boundaries are reached. It is because they
provide an unbounded path that only verbs of motion occur in this
context.
There is another class of interpretations of durch that has to be analysed
in the same way. Up to now I have claimed that, except for mass nouns,
the condition of complete traversal must hold. This however is not always
true (again) in the case of verbs of motion. Often there are two possible
interpretations for the examples in (34).
(34)
a. Er geht durch den

He walks through
b. Er fhrt durch die
He drives through
Park.
the park,
Stadt.
the city.
On one interpretation the condition of complete traversal holds, on the

other it does not. In the latter interpretation, the located object only
moves around inside the relevant region, as in the case of mass nouns. I
think one can deal with this interpretation in the same way as with the
mass problem. First, note that not only masses can be individuated and
conceptualized as bounded objects. Objects can also be viewed as masses
if their boundaries are ignored. That is, we can look at objects from the
outside and conceptualize them as bounded objects within a larger context; or we can look at them from the inside and conceptualize them as
objects extending beyond the context, and therefore as masses.
If we choose the first alternative in interpreting the sentences in (34),
we get the reading in which the condition of complete traversal is fulfilled.
If we choose the second one, we get the mass interpretation. While in
the first case we get an accomplishment structure, that is, an extended
but finite situation, in the second case we get an unbounded process. The
borders of the region that have to be reached in order to terminate the
situation have not yet been reached.
The last case to be dealt with is the one involving a two-dimensional
object of reference as illustrated in configuration (c) in Figure 1. First
the fact why these cases are not blocked by the dimensionality constraint
discussed in 2.1. must be explained. There it was argued that durch
imposes the same dimensionality constraints as in does. The examples in
(35) show that this assumption still holds.
(35)
a. der Blick durch das Fenster

the view through the window
243
a', die Pflanze im Fenster

the plant in the window
b. Er geht durch die Tr.
He walks through the door,
b'. Er steht in der Tr.
He stands in the door.
Obviously the relevant dimensions for inclusion are given by the vertical
and the horizontal axes. The third dimension, that can be characterized
as the observer axis (cf. Lang 1987), is irrelevant with respect to this
relation. I will not try to examine why this should be so. For some reason
the vertical axis is more important than the other two. Perhaps it is
possible for two-dimensional objects that extend in this direction to be
reconceptualized as three-dimensional. If this is the case, the examples in
(35) receive the standard interpretation. Otherwise, the part of the path
included in the object of reference reduces to a point. Since in contrast
to the objects extending only horizontally (market square, pavement),
most of the two-dimensional objects that extend vertically {window, door,
banner) can be easily viewed as objects independent of their surroundings,
I prefer the first solution. Due to the fact that the third dimension of
these objects only has a minimal extension, in combination with durch
the path extends beyond the object.
Having found a way of dealing with the apparently inhomogeneous
readings of durch, a semantic form that can account for all of the
interpretations presented above will be proposed in the next section.
5. The semantic representation of durch

As I said in 2.1. the semantic representation of durch contains the same
inclusion component as in [-Dir], The condition on the dimensional
properties required of the object of reference is due to this component.
(36)
(LOC (x, INT (y)))
To deal with the component in which the path is related to the reference
object, an intersection line S (y) relative to the object of reference is
constructed as follows:
Let a and b be two distinct boundary points of the object of reference
y such that S (y) can be defined as the interval [a, b] and thus intersects
y. Except for a and b, S (y) consists of inner points of y only.
244
Ingrid Kaufmann
Which part of S (y) is actually present in the context depends on how

the object of reference is conceptualized. If it can be individuated from
the outside, that is as a bounded object, the whole interval [a, b] is part
of the context. If the object of reference is looked at from the inside, that
is, if its border is not part of the context, only the open interval (a, b) is
accessible. Sc (y) is that part of the interval [a, b] that is actually present
in a context c.
To deal with the relation between D (x) and S (y), the inclusion relation
INCL is defined. This relation is to be interpreted as follows:
(37)
INCL (D (x), S (y)) is interpreted as D (x)

S (y) iff Sc (y)
= [a, b];
INCL (D (x), S (y)) is interpreted as D (x) = S (y) iff Sc (y)
= a, b).
durch can now be represented as follows:

(38)
* (INCL (D (x), S (y)) & LOC (x, INT (y)))
Since deformation of paths (cf. Habel 1989: 54) is always possible, this
representation allows one to deal with the conceptually based effect that
shows up in the interpretation of phrases in which a bounded object is
conceptualized from the inside (cf. (39)).
(39)
Er lief stundenlang durch den Wald.

He walked through the city for hours.
Sentences like this usually get the interpretation that the route of the
moving object is not straight, but somehow winding. This can be explained by an interaction between the semantic information that requires
a path approaching the (unreachable) boundary point b, and the knowledge that since the object is bounded somewhere, by taking a straight
route the borders must be reachd sooner or later. Since on this reading
the borders are not reached during the relevant situation, one chooses a
deformation of D (x) that does not approach the borders directly.
Finally I want to show how composition of complex phrases is carried
out by discussing one occurrence of durch in more detail. In the representations below, only the information that is necessary for composition
is given.
(40)
durch den Wald wandern

wander through the wood
245
wandern:
conceptual information: movex: > S
semantic information: (MOVE () & Wander () & ()) (s)
Assuming to be time and S the set of spatial regions, conceptually
MOVE (x) has to be interpreted as a continuous mapping from into
S, assigning to every period of time the region in which is located at
that time. Thus a structured path is built up. The semantic representation
of wandern includes the theta-grid consisting of the predicate variable P,
the variable for the external argument x, and the variable for the situational argument s.
Wald\
conceptual information: <(a, b), vert)
semantic information: (Wald ())
durch den Wald wandern:
(41)
Xs (MOVE () & WANDER () & INCL (D (), S (Df

(WALD))) & LOC (, INT (Df (WALD)))) (s)
" D f " is to be interpreted as the definiteness operator. (41) is the complex

semantic representation of the phrase in (40). To obtain the right interpretation further conceptual information is necessary. The correct interpretation of durch proceeds as follows:
In order to fulfill the requirements of the location component, the
dimensionality restrictions must be met. Moving objects are usually threedimensional, therefore the object denoted by the internal argument must
have at least three dimensions as well. The Object Schema of Wald
provides the information that this is in fact the case. The information
encoded by the Object Schema is the following: Wald is a bounded object
(this is encoded by the angle brackets < )). In the case of Wald it could
be reasonable to assume that boundedness results from the definite
determiner. As in the case of mass nouns, by means of definitness a
particular wood is isolated. I will not deal with this problem here. What
is relevant is only that boundedness should be somehow given, "a", "b"
and "vert" label the three object axes. While "vert" is bound to the
vertical axis, there is no specification for "a" and "b". The brackets
connecting a and b show that they are integrated axes, that is, they define
two dimensions. These dimensions are not conceptually discriminated as,
for example, in the case of a brick. Since the third axis is bound to the
vertical axis, a and b must refer to the horizontal level.
246
Ingrid Kaufmann
Now this information is important to determine the course of D (x).

The representation of durch only gives the information that it must include
a line intersecting the wood. From the conceptual interpretation of the
verbal predicate WANDER (which is not made explicit here) one knows
that movement takes place on the ground. D (x) has to be specified by
the path constructed by move, so we know that S (y) cannot coincide
with the vertical axis but only with one of the horizontal axes; which one
remains open.
This information is sufficient for an interpretation of the phrase. The
entity denoted by the grammatical subject of the phrase, no matter what
it is, is located on the ground inside the wood and is walking along from
one border to the other. If Wald is conceptualized as unbounded, the
object would move along an equally unbounded path inside the wood.
6. Summary
The interpretation of the preposition durch proceeds as follows: The
semantic representation delivers two components of meaning: an inclusion
component, which places the condition on the dimensional properties of
the object of reference; and a path component whose central element is
the parameter D (x). D (x) must improperly contain an interval S (y)
which intersects the object of reference y. Conceptual information determines the conditions. This conceptual specification makes use of the
interactions between the located object and the object of reference.
Notes
1. This paper is a revised version of an article that appeared in Kognitionswissenschaft 1
(1990). It is based on a talk given at a workshop on the semantics of prepositions in
Saarbrcken, February 1990. It grew out of the DFG-Project on spatial prepositions
W U 86/10.
References
Bierwisch, Manfred
1983
"Semantische und konzeptuelle Information lexikalischer Einheiten", in: R.
Ruzizka W. Mtsch (eds.), Untersuchungen zur Semantik, 61-99. Berlin:
Akademie-Verlag.
247
1988)
"On the grammar of local prepositions", in: M. Bierwisch W. Mtsch

I. Zimmermann (eds.), Syntax, Semantik und das Lexikon, 1-65. Berlin:
Akademie-Verlag.
Habel, Christopher
1989
"Zwischen-Bericht", in: C. Habel - M. Herweg - K. Rehkmper (eds.),
Raumkonzepte in Verstehensprozessen, 37-69. Tbingen: Niemeyer.
Herweg, Michael
1988
Zur Semantik einiger Prpositionen des Deutschen. berlegungen zur Theorie
der lexikalischen Semantik am Beispiel von in, an, bei und auf. Lilog-Report
21.
Kaufmann, Ingrid
1989
"Direktionale Prpositionen", in: C. Habel M. Herweg K. Rehkmper
(eds.), Raumkonzepte in Verstehensprozessen, 128-149. Tbingen: Niemeyer.
1990
"Semantische und konzeptuelle Aspekte der Weg-Prposition durch", in:
Kognitionswissenschaft 1; 15-26. Springer.
Lang, Ewald
1987
"Semantik der Dimensionsauszeichnung rumlicher Objekte", in: M. Bierwisch E. Lang (eds.), Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven, 287-458. Berlin: Akademie-Verlag.
1990
"Primary perceptual space and inherent proportion schema: Two interacting
categorization grids underlying the conceptualization of spacial objects", in:
Journal of Semantics 7; 121-141.
Wunderlich, Dieter
1991
"How do prepositional phrases fit into compositional syntax and semantics?",
in: Linguistics 29: 591-621.
The meaning of German projective prepositions:

A two-level approach*
Ewald
Lang
Introduction
Prepositions seem to be notoriously unruly. The conclusions of a recent
monograph on the semantics of prepositions start with the complaint:
"Attempts to give an account of lexical meaning remain the weakest link
in artificial intelligence and logical semantics. Neither discipline has ever
offered lexical representations that satisfactorily account for speakers'
uses; both have discovered that lexical meanings simply cannot be regimented into neat bundles of necessary and sufficient conditions" (Herskovits 1986: 193).
While one may fully agree with the first part of this statement as
concerns the actual state-of-the-art, one may be less pessimistic regarding
the claim that lexical meanings are resistant to a clear-cut analysis on
principled grounds. Annette Herskovits presents an impressive picture of
the difficulties any semantic account of prepositions has to cope with.
However, one may fully appreciate the meticulous descriptions she provides without being forced to adopt her premises and/or conclusions.
There is room, I think, to put the analysis of prepositions on a different
basis. The two-level approach proposed in what follows is an attempt in
that direction. Before getting down to the basics, a few introductory
remarks might be in order.
0.1. Background
The two-level approach to the semantics of projective prepositions outlined below is an extension of a framework which was originally developed
in a comprehensive analysis of dimensional expressions (see Bierwisch
Lang 1987, 1989; Lang 1988, 1989 a, b, 1990, 1991b, 1992 Lang et al.
1991; Simmons 1991, 1992) and, in the meantime, has been adopted and
modified by other researchers in a series of studies on prepositions
(Herweg 1988; Wunderlich Herweg 1991; Wunderlich Kaufmann
250
Ewald Lang
1990, Bierwisch 1989; Kaufmann 1989, 1990, Habel 1989, Li 1991),

position verbs (Maienborn 1989 a, b, 1991), route descriptions (Carstensen 1991) to name but a few. The move from dimensional terms to
prepositions involves two aspects of continuity to be noted: one concerns
the conceptual domain shared by the lexical items at issue, the other
relates to the types of semantic problems to be coped with.
0.2. Spatial knowledge

The inherent affinity between spatial expressions designating the height,
width or depth of objects and prepositions locating objects in relation to
the top-bottom, front-back etc. axis of other objects is intuitively obvious.
We stay within the same domain, so the extension suggests itself. But it
makes a difference wherefrom you start. Thus, the attempt to use the
basic concepts underlying the dimensional designation of objects for an
analysis that aims to account for the localization of objects in space is
tantamount to claiming that the inventory involved in the former is more
basic than the one involved in the latter.
In other words, an account of local prepositions which essentially
draws on descriptive means that are independently justified by analyses
of dimensional terms, verbs of position and other spatial vocabulary
should be preferred to an account which draws solely on concepts and
rules that are specifically devised for the analysis of spatial prepositions
(as holds for, e. g., Herskovits' work).
0.3. Like phenomena

The problems of polysemy raised by projective prepositions like above,
in front of, behind etc. are strikingly similar to the ones we face regarding
the range of interpretations assignable to high, wide, deep etc. Thus we
would expect the representational means that account e. g., for the difference between primary designation (high building, tall tower, deep hole
*high/*tall/*deep ball) and contextual specification (deep plot of ground,
tall pole) also to be appropriate to cope with the difference between
intrinsic vs. extrinsic uses of projective prepositions (cf. the poster above/
in front of the building vs. the ball is above/in front of the pole). In short,
German projective prepositions: A two-level approach
251
spatial prepositions make a rewarding test case of the framework that

proved to be working for dimensional adjectives and other spatial vocabulary.
0.4. Points of divergence

The approach I am going to propose would best be introduced against
the background of Herskovits' work. But instead of reviewing it in detail,
which no doubt remains a rewarding task, I will confine myself to
highlighting some fundamental points of divergence. The two-level approach to the semantics of prepositions to be outlined in this paper differs
from the approach taken by Herkovits (or by related work within a
prototype-semantics framework, e.g., Vandeloise 1984, Langacker 1986,
Hottenroth 1986, Lakoff 1987) in some respects which, if discussed in
full detail, would take us back to the very fundamentals of semantics.
Within the limits of this paper I will touch them briefly in Section 3
below while sketching the alternative assumptions that form the basis of
the two-level approach.
0.5. Formalization
The approach forwarded here also entails a somewhat different attitude
toward the role of formalization in linguistic theorizing. While Herskovits,
with reference to "the subtlety and complexity of language use" (1986:
19), does not see any advantage in formal representation or implementation, I take the stand that implementing a (sufficiently formalized)
linguistic analysis may prove to be a very useful means to check up the
consistency and exhaustiveness of the underlying theory. Thus, the linguistic analysis presented in Lang (1988, 1989 a, b) has been implemented
as a Prolog prototype called OSKAR (Lang - Carstensen 1989, 1990;
Lang et al. 1991).
What good was that? We have gained a lot from the feed-back between
object theory and program that happened to emerge in the course of
implementation. Besides minor corrections and a couple of relevant
improvements effected by this feed-back, the most interesting result is
the non-arbitrary correspondence between the modular structure of the
theory to be implemented and the building blocks of the program implementing it (cf. Section 3 below).
252
Ewald Lang
1.
Projective prepositions 1
1.1. Scope of the data

German locative expressions include a subset of projective prepositionals,
the inventory of which is formed by six basic prepositions:
(1)
a. ber-unter;
b. vor-hinter;
c. rechts links
and a somewhat idiosyncratic set of adverbials somehow derived from

them: 2
(2)
a. darber-darunter, oben-unten,
oberhalb-unterhalb
b. davor-dahinter,
vorn-hinten
c. rechts-links davon, zur Rechten-Linken von, rechterhand-linkerhand
Needless to say, it remains a major task to elucidate the interplay of

morpho-syntactic and semantic properties in this field. However, as I will
be concerned with semantic aspects of these expressions that seem to be
constant across categorial distinctions, I will confine the analysis to the
core prepositions in (1). Moreover, I will deal only with the locational
meanings of these prepositions, restrict the data to attributive constructions or simple copula sentences as in (3), and neglect the distinction
between local vs. directional interpretation. 3
1.2.
Undisputed preliminaries
1.2.1. The general structure of local prepositions

Semantically, phrasal or sentential expressions of the form:
(3)
(ist) bei/an/auf/ber)vor/rechts von ...

y
[x (is) near/at/on/above/in front of/to the right of y]
all involve the localization of (the entity to be located, for short: LE)
with respect to a spatial region surrounding y (the reference object, for
short: RO). This much is common sense among all students of local
prepositions and we will adopt this view by assuming (4) as a first
253
approximation to representing the lexical meaning of local prepositions.

Leaving out controversial details of competing recent proposals (Herweg
1988; Wunderlich - Herweg 1991; Bierwisch 1988; Kaufmann 1988,
Habel 1989), the general format to represent the lexical meaning of spatial
prepositions is supposed to be this:
(4)
y A x (LOC (x, R E G * (y)) ...)
In this schema, "x" and "y" are variables to be replaced with appropriate
representations of (object-)individuals; "LOC (x, R)" is a localization
relation to be read as "the place of is locally included in some region
R"; " R " in turn is determined by a set of functions REG*, the elements
of which assign a preposition-specific region to the object represented by
"y". The " . . . " mark the place for the possible addition of further constraining conditions.
1.2.2. Theta-roles
Mediated by the notion of theta-roles, the semantic distinction of RO
and LE correlates with the syntactic distinction of internal vs. external
arguments associated with the preposition at issue. Thus, in terms of
syntactic-semantic composition, the argument y occupying the governed
position ( = the one assigned case by the prepositional head governing
it) is assigned an internal theta-role A y to be saturated by a constituent
naming the reference object RO. The argument is assigned the external
theta-role A x, which is discharged depending on whether the given PP
syntactically functions as an argument, a predicate, or a modificational
adjunct (cf. Bierwisch 1988). In our sample of data, the external thetarole is eventually saturated by a constituent naming the located object
LE. Consideration of the syntactic roots of the distinction RO vs. LE is
relevant in deciding on transitivity and conversion relations with projective prepositions. For details on theta-roles of prepositions see the discussion in Rauh (1991 a, 1991 b), Haumann (1992).
1.2.3. Delimitation of meaning

What is specific to each of the prepositions in (1) and (3) is the way it
determines a particular neighbourhood region of y. Thus, as a first step
in specifying REG* in (4), it makes sense to distinguish a BEI-region, an
AUF-region, an BER-region etc. Given this, the next step is to establish
254
Ewald Lang
which of these particular regions surrounding y are disjoint and which

possibly overlap. Being determined by the putative prime PROXIMITY
solely, the BEI-region can be considered to be the least specific of all
preposition-based neighbourhood regions (which is obviously somehow
related to the fact that bei, in contrast to the items in (1), neither has a
lexical opposite nor takes measure phrases).
Now, in a sense, the projective prepositions in (1), which come in three
pairs of lexical polar opposites, seem to divide the BEI-region into six
(partially overlapping) subregions. This much is opinio communis among
students of prepositions, but there is much debate on the conceptual basis
as well as on the proper semantic ingredients by means of which the
prepositions can be accounted for. Instead of reviewing all the relevant
literature (for a useful survey see Retz-Schmidt 1988), in the section to
follow I will give an outline of my own proposal. Put in a nutshell, the
claims I am going to discuss amount to the following:
(5 a) For the prepositions in (1), the component R E G * in (4) can be
specified by a set of semantic parameters that determine the pertinent
neighbourhood regions by making reference to conceptually available
axis-based sides ("top", "bottom", "front", "back" etc.) of the RO.
(5 b) The parameters for side-assignment are derived from the three axes
that form man's primary perceptual space, viz. the Vertical, the Observer
Axis, and the Horizontal. It is in relation to these axes that objects are
assigned dimensions, positions, and locations in space and it is the
properties of these axes that govern the semantics of projective prepositions.
(5 c) The relevant spatial properties of objects are to be represented on
a level distinct from that of (4) by means of Object Schemata that reflect
our everyday knowledge of spatial objects (which is compatible with, but
distinct from, our linguistic knowledge needed to use object names properly).
(5d) The need to account for the various intrinsic vs. extrinsic uses of
these prepositions as well as for relations like transitivity and conversion,
is what calls for a two-level approach as proposed here.
The move from dimensional terms to prepositions involves two aspects
of continuity worth being noted. The first, discussed in 2.1., concerns the
shared spatial knowledge forming the conceptual domain which both
groups of lexical items refer to. The second, discussed in Section 3., relates
to the fact that, to a certain extent, projective prepositions and dimension
expressions pose the same types of semantic problems.
2.
255
Basic assumptions:
The framework of spatial knowledge
2.1.0. What are the basics?

It is intuitively obvious that there is an inherent affinity between spatial
expressions designating the height, width or depth of objects and prepositions locating objects in relation to the top-bottom, front-back etc.
axis of other objects. We stay within the same domain, so the extension
suggests itself. But it makes a difference where you start from. Thus, an
approach that draws on basic concepts that proved adequate for the
analysis of dimensional terms, verbs of position and other spatial vocabulary should be preferred to an account which draws solely on concepts
and rules that are specifically devised for the analysis of spatial prepositions (as holds for, e.g., Herskovits' work).
As regards the source of spatial knowledge and the inventory of
conceptual elements that are made use of in the descriptive framework,
the "experiential" approach adopted by Herskovits and others is, roughly
speaking, observer-centered and situation-based. That is, the primary
concepts for approaching the meaning of locative expressions are defined
in terms of prototypical situation types which involve a human being in
"canonical position" (standing upright on horizontal ground) as the
central source for orientation cues (among them the six half-line axes
relevant for the description of projective prepositions (see Herskovits
(1986: 156 ff.)).
In contrast, the two-level approach I am advocating may be called
object-centered and axes-based, as it primarily draws on an inventory of
categorized object concepts (see Section 4.) and the categorization grids
that define their constitutive spatial properties. In this approach, then,
spatial knowledge is basically organized by what I call the Primary
Perceptual Space (PPS), that is, a system of axes that form our internal
model of the external space. (Though it is difficult to compare the
ontologies of both approaches, I dare say the latter is more general and
comprehensive, as it seems to obey Occam's razor more clearly than
Herskovits' system.)
Presupposing the details discussed at length in Lang (1987, 1989), Lang
Carstensen (1990), we will take a brief look at these axes by examining
briefly how they are involved in the semantics of dimensional terms and
projective prepositions.
256
Ewald Lang
2.1.1. The Vertical

Originating from the effects of gravitation as perceived by the human
body, the Vertical is an orientation cue which is ubiquitous and constant,
that is, available everywhere in our terrestrial habitat and with the same
effect at all times. This makes verticality a primary cue of spatial perception to the effect that, within Primary Perceptual Space, the Vertical is
granted the status of an independent axis.
Man's upright posture assigns the Vertical a natural foot on the earth's
surface (or on a parallel plane that serves as ground level) and a fixed
direction which is determined as geofugal by the top-bottom asymmetry
of the human body. In geometrical terms, the Vertical thus has the
attributes of a ray or directed segment. The foot serves as a zero-point
F such that any points F', F" (distinct from F) on the Vertical define
directed segments that increase unidirectionally. The direction defined by
F, F', F" ... etc. corresponds to the direction of the height scale whose
intervals are construed as increasing degrees of height assigned to the
objects in question.
Due to these features, the Vertical is physically as well as conceptually
the most salient and also the dominant axis of Primary Perceptual Space;
the other axes are defined in relation to the Vertical. In other words, the
Vertical is the backbone of spatial orientation.
2.1.2. Linguistic reference to the Vertical

Consider the sample of spatial expressions in (6) showing the inherent
relationship of dimension adjectives (in dimensional as well as in distance
or positional interpretation) and projective prepositions.
(6)
a. Das Fenster ist 2 m hoch, (dimension and distance interpretation)

'The window is 2 metres high.'
b. Das Fenster ist 2 m hher als der Tisch, (distance interpretation)
'The window is 2 metres higher than the table.'
c. Das Fenster ist 2 m ber dem Tisch, (projective preposition)
'The window is 2 metres above the table.'
The semantic differences to be correlated with the morphosyntactic distinctions between dimension adjective, distance adjective, 4 and projective
preposition emerge from the different ways in which the points F, F', F"
Figure 1. Interpretations of Das Fenster

ist 2 m hoch
257
Figure 2. Spatial configuration referred to

by (6 b) and (6 c)
of the Vertical are instantiated by an object said to be high (in itself) or

to be located in a higher place (compared to another object) or to be
located in the region determined by the highest part ( = the top side of)
the reference object. In a way, the distance adjective forms the bridge
between dimension term and projective preposition. Figure 1 below illustrates the two interpretations of (6 a), Figure 2 indicates those elements
that the interpretations of (6 b) and (6 c) have in common, their differences
will be commented on.
The Dimension Interpretation of hoch/high in (6 a) involves the projection of two axis endpoints E, and E2 onto the Vertical such that the
segment formed by E,E 2 is transformed into a directed segment defined
by the points F'F" of the Vertical. The deciding condition of this interpretation is that the relevant points of the Vertical, i. e., F', F", are
instantiated by axial endpoints of one and the same object x. Given this,
it it these endpoints that define the object's bottom and top, respectively.
In contrast, the Distance Interpretation of hoch/high in (6 a) is based
on the condition that the points F', F" of the Vertical are instantiated by
distinct objects. In the example given it is the window as a whole which
is projected onto F', thereby defining the directed segment FF' of the
Vertical as a distance between ground level (the default value of F) and
window (instantiating F'). So the comparative construction in (6 b) involves two distances to be compared, the directed segments marked as
FF'and F j F / i n Figure 2, which have the foot F in common (F = Fj).
In short, the Dimension Interpretation is based on a directed segment
defined within one object, which entails the object being assigned bottom
258
Ewald Lang
and top; whereas the Distance Interpretation is based on a directed

segment defined between objects that are conceived as mere points, which
entails that the objects' sides (including top, bottom, front and rear) are
disregarded.
The interpretation of the projective preposition in (6 c) involves aspects
of both dimension and distance interpretation plus something special.
The dimension aspect is involved indirectly in establishing the top of the
table, which, in turn, determines the relevant BER-region (dotted area
in Figure 2). The distance aspect is involved inasmuch as the local relation
between the located object and the reference object is conceived as a
measurable distance this is the basis for the near-equivalence of (6 b)
and (6 c).
To sum up, the lexical meanings of 2 m hoch, 2 m hher and 2 m ber
all involve the Vertical to localize objects in space; but while the adjectives
operate on geometric concepts like axes, directed segments, (end-)points,
sides etc., the projective preposition ber (or any other item of (1) or (2))
involves a topological concept of neighbourhood region as indicated by
the dummy REG* in (4). Semantically, the projective prepositions (unlike
the spatial adjectives, on the one hand, and unlike the pure topological
prepositions such as bei or an, on the other hand) thus form a field of
structure formation where geometry and topology intersect and interact.
The different though obviously related ways in which reference to the
Vertical is involved in (6 a-6 c) are rooted in the internal semantic structure
and the morphosyntactic features of the spatial expressions at issue. What
we are looking for, then, is a representational format capable of capturing
both the common reference to the Vertical and the different ways in
which this reference is grammatically coded. With this end in view I will
assume lexical entries for hoch and ber as shown in (7) that associate
the morphosyntactic features of these lexical items with semantic representations (designed as Semantic Forms along the lines of Bierwisch
Lang 1987, 1989 and Bierwisch 1988-see 3.2. below) that account for
dimension adjectives (7 a), distance adjectives (7 b), and projective prepositions (7 c) respectively:
(7)
a. /hoch/; [ + N, +V]; A c A x [[QUANT VERT x] 2 [v + c]]

/niedrig/; [ + N, +V]; A c A x [[QUANT VERT x] c [ v - c]]
b. /hoch/; [ + N, +V]; A c A x [[QUANT VERT [DIST y x]] 2
[v + c]]
c. /ber/; [ - N , - V ] ; A y A x [Loc [Loc [VERT*F' y]]]
/unter/; [ - N , - V ] ; A y A x [Loc c [Loc [VERT*F y]]]
259
The central point is how the semantic parameters that refer to the Vertical
are integrated into complex units that differ as to argument structure
and syntactic categorization. I cannot repeat here the details that support
the design of the representations given above, so let me note just three
points that are of immediate interest to our topic.
First, note that compared to the dimension adjectives in (7 a), the
distance adjective hoch in (7 b) embodies an additional component DIST,
which induces a third argument place y, where y figures as a parameter
to be replaced with an appropriate value from the context. Thus, in the
interpretations of (6 a) and (b), this parameter y will assume the value
"ground level", that is, the default value to instantiate F (see Figure 1),
where F is brought in by the semantic component VERT. But y can also
be replaced with a value available from the structural context as in, e. g.,
das Fenster ist vom Fuboden aus 2 m hher als der Tisch.
Second, note that there are differences in antonymy relations. The
polar antonymy of the dimension terms hoch-niedrig draws on scaling
operations (represented by " 2 ... + " and
... " in (7a)) that are
reversals of each other but operate on the same scale determined by
VERT. In others words, the Vertical yields the common scale basis on
which opposite operations underlying hoch [high] and niedrig [low] are
carried out. This is reflected in (7 a) by according the parameter VERT
the same place in the semantic form of both adjectives.
The distance adjective hoch in (7 b) does not have an antonymous
counterpart niedrig (note that, e.g., Der Tisch ist 2m niedriger als das
Fenster is not a converse of (6 b)). In some cases, this gap can be filled
by using the distance adjective tief as in Davos ist hher gelegen als Zrich
= Zrich is tiefer gelegen als Davos [D is higher-lying than = is
lower-lying that D]. 5 However, despite the fact that, in certain contextual
frames, hoch and tief may denote opposite directions, they are not lexical
antonyms (for details cf. Lang 1989). These observations are accounted
for in (7 b) by assuming a parameter DIST (a component shared by all
distance terms like hoch, tief weit and their derivatives) such that the
particular representations of hoch or tief are obtained by combining DIST
with a specificatory component VERT or OBS. Though the so-called
distance adjectives are poorly understood as yet, the representation suggested in (7 b) seems to capture at least some of their peculiarities.
At any rate, assuming a separate component DIST that induces the
additional argument y will account for the fact that the items containing
it refer to distances defined between objects, while allowing for DIST to
260
Ewald Lang
be specified by VERT in the manner just outlined will account for the
fact that the distance adjective hoch refers to the Vertical as the carrier
axis of the distance it denotes.
The antonymy of the prepositions ber-unter emerges from applying
a relation of local inclusion (represented by 'Loc ... ^ Loc ... y') to
neighbourhood regions of the reference object y that are determined by
opposite sides of y. These object sides are defined by coincidence with
the designated points F, F' of the Vertical induced by VERT. In other
words, the Vertical yields the orientation cue with reference to which an
object y can be assigned opposite sides which, in turn, define correspondingly opposite neighbourhood regions of y. Thus, the antonymy of berunter is due to their making reference to neighbourhood regions of y
determined by opposite sides of y. This is reflected in (7 c) by assuming
two correspondingly distinct parameters, viz. VERT*F' and VERT*F,
which are taken to represent sets of conditions for identifying the respective regions. Note that F, F are metavariables ranging over the points
F, F', F" of the Vertical induced by VERT in such a way that F refers
to the object's bottom as defined by the actual foot F a , whereas F' refers
to the object's top as defined by the corresponding F' a .
I hasten to add that the labels used to symbolize the relevant parameters
in (7a)-(7c) are meant to give an idea of their internal composition.
What really matters is the substantial interpretation these parameters are
given as notional elements within a theory. As regards VERT and DIST,
this has been done at length in the studies quoted above. Now, based on
the assumption that the representations suggested in (7 c) can be shown
to be a notational variant of the format used in (4), we will consider
VERT*F' and VERT*F promising candidates to replace the place-holder
REG* in (4), yielding:
(8)
a. ber. A y A x (LOC (x, VERT*F' (y)) ...)

b. unter. A y A x (LOC (x, VERT*F (y)) ...)
Spelling out the full range of a semantic parameter is a thorny business

to do, so we will focus on answering the simple question: What's in a
VERT*F?
2.2.1. The Observer Axis
Originating in the visual organ, this axis is determined by the line of sight
of a (potential or actual) observer. Because of this it is in contrast to
the Vertical's ubiquity and constancy flexible in two respects. First,
261
having a movable source (i. e., the observer), this axis does not have a
fixed anchorage in the surrounding physical space but is induced by a
moving human interpreter of physical space. Second, the Observer Axis
has an anatomically determined pivot allowing for a 180 turn in either
of two planes (vertically and horizontally). Being the axis of depth
perception, the Observer Axis certainly provides another pillar of spatial
perception. However, due to the fact that the Observer Axis is bound to
human carriers whose normal position (upright posture) is defined as
normal only in relation to the surrounding physical space, this axis is not
as independent an axis within Primary Perceptual Space as is the Vertical.
Defined by a (real or imaginary) observer's gaze, the Observer Axis
has a direction which is biologically fixed as away from the observer, and
it has a bounding point O, which (disregarding disparity) is given by the
anatomically determined origin of an observer's line of sight.
In geometrical terms, the Observer Axis can also be described as a ray
or directed segment. It is defined by the bounding point and has a
direction in that any points O', O" determine unidirectionally increasing
distances from O. Due to this, the Observer Axis provides the scale basis
for DEPTH ASSIGNMENT. Originating in the visual organ of a human
observer, however, the Observer Axis of Primary Perceptual Space simultaneously serves other purposes as well. It also forms the basis for
perspectivizing spatial objects.
PERSPECTIVIZATION consists of assigning objects observer-determined
"fronts" and "backs" etc. and locating objects x, y relative to one another,
as in is in front of y or y is behind etc. In short, the Observer Axis is
the carrier of two major cognitive cures of spatial orientation: DEPTH
ASSIGNMENT and PERSPECTIVIZATION. Coming from the same biological
source, these two cognitive devices closely interact. Note, however, that
DEPTH ASSIGNMENT and PERSPECTIVIZATION are partially independent in
that they draw on distinct spatial structures; therefore, they have to be
kept separately (cf. 2.2.2. below).
In the unmarked case, given by the position of the eyes of an observer
in upright posture, the Observer Axis is orthogonal (at 90) to the Vertical.
In the other relevant configuration, the Vertical and the Observer Axis
lie at an angle of 180 such that they run parallel but in diametrically
opposed directions. The third case, where the Vertical and the Observer
Axis run in the same direction (at 0), is perceptually quite clear but,
interestingly enough, does not constitute a conceptually relevant parameter (cf. Lang 1987, 1989).
262
Ewald Lang
2.2.2. Linguistic reference to the Observer Axis

The examples (9 a)-(9 d) show, in a way quite parallel to those in (6), the
inherent relationship of dimension terms, distance terms, and projective
prepositions. But due to the two functions of the Observer Axis, i.e.,
DEPTH ASSIGNMENT and PERSPECTiviZATiON, we have to consider a wider
range of possibilities.
(9)
a. Die Kiste ist 2 m tief.

(dimension interpretation)
'The box is 2 metres deep/in depth.'
b. Die Kugel ist 2 m tief im Mauerwerk, (distance interpretation)
'The ball is 2 metres deep in the masonry.'
c. Die Kugel ist tiefer im Korb als das Dreieck, (distance interpretation)
'The ball is deeper in the basket than the triangle.'
d. Im Korb ist das Dreieck vor der Kugel.
(projective prepositions)
Im Korb ist die Kugel hinter dem Dreieck.
'In the basket, the triangle is in front/behind of the ball.'
Figure 3 illustrates spatial configurations to which the sentences in (9)

may be applied. In these cases, the relevant points of the Observer Axis,
that is the bounding point and its successors O', O", can be furnished
with the appropriate instantiations needed for DEPTH ASSIGNMENT.
As regards the semantics of the prepositions vor-hinter, however, DEPTH
ASSIGNMENT is only part of the story. The close relationship between (9 c)
O'
O"
s
3
-
Dimension
'
"
'
"
'
Distance
Distance /
projective preposition
Figure 3. Spatial configuration referred to by (9 a) and (9 d)
263
and (9d) rests on the availability of a hollow object (viz. the basket)
forming the frame of reference with respect to which the triangle and the
ball can be ordered on the depth axis. If you imagine the scene with the
basket removed, you will immediately realize that the spatial relation
between the triangle and the ball can no longer be described by tief(er),
but has to be referred to by means of the projective prepositions vorhinter. This is the point where PERSPECTIVIZATION takes over.
There are basically two strategies for assigning observer-related sides
to an object which does not have any intrinsically fixed front-back
distinction. Each of these strategies corresponds to a specific spatial
situation defined by the observer's location with respect to the object(s)
at issue. Following familiar terminology (e.g., Herskovits 1986: Chapter
10), one is called the ENCOUNTER SITUATION the observer faces the
object from a distance; the other is called the COINCIDENCE SITUATION
the observer transfers his own front-back distinction onto the object in
question. Consider the situations according to which an object, say, a
block, is assigned "front" (F) and "back" (B), and how, depending on
that, a ball and a triangle are located in relation to the block. Figure 4
shows the relevant spatial configurations of observer, objects, and observer-related object sides.
While the bounding point of the Observer Axis is given by the
observer in a like manner for both situations, there is an important
difference between the two situations as to the role of the successor points
ENCOUNTER SITUATION
COINCIDENCE SITUATION
O'
264
Ewald Lang
', ". Consider the interpretations of ist vor der Kiste/z ist hinter der
Kiste 'x is in front of the box/z is behind the box', where die Kiste 'the
box' is the reference object with respect to the side-based neighbourhood
regions of which the objects and are located.
In the encounter situation, point O' is instantiated by the less distant
object face, which is then designated as "front" (F) or "accessible face";
O" is instantiated by the more distant object face, which is designated as
"back" (B) or "remote face". Note the analogy to the Dimension Interpretation of deep in Figure 3 above. According to this strategy, "front"
and "back" are assigned to an object in terms of increasing distance from
on the Observer Axis. This is the source of the well-known mirror
effect of assigning an object an observer-determined front (i. e., the front
of an object is the side which the observer faces or, which amounts to
the same, the side which faces the observer). Given this, we have the
appropriate side-determined regions (surrounding the block) at our disposal, with reference to which the ball is said to be vor 'in front o f , and
the triangle to be hinter 'behind', the block. The crucial feature of this
strategy of assigning observer-related sides and side-related locations to
objects lies in the fact that the concepts "front" and "back" consistently
correlate with less distant and more distant object faces, respectively.
In the COINCIDENCE SITUATION, however, the crucial point is that the
reference object to which sides are assigned (i. e., the box), coincides with
the observer her/himself in instantiating the bounding point (cf. the
lower part of Figure 4). According to this strategy, then, the assignment
to objects of observer-related "fronts" and "backs" is not a projection of
object sides onto points O', O" of the Observer Axis, but a transfer of
the observer's intrinsic front and back onto the object in question
hence the altered distribution of "front" and "back" with respect to the
block in Figure 4. In fact, based on COINCIDENCE the triangle may be
said to be vor 'in front of and the ball to be hinter 'behind' the box.
Even under this view, however, the Observer Axis remains available for
localizing any objects that may instantiate O' outside the reference object,
though, of course, this is restricted to front-based regions.
As regards the semantic parameters involved in vor and hinter, let me
summarize the points made so far:
First, the observation that vor and hinter, with respect to the configuration of objects shown in Figure 4, may assume the full range of
interpretations assignable by either strategy (i. e., ENCOUNTER and COINCIDENCE) is based on the fact that none of the objects involved (box, ball,
265
triangle) has an intrinsic front or rear side. Hence, objects like these can
be assigned fronts and backs only via projection from an external source.
Second, the observer, who forms the source of these projected assignments, has two orientation cues at her/his disposal, i. e., DEPTH ASSIGNMENT and PERSPECTIVIZATION, either of which may be used to assign fronts
and backs to objects. Hence the range of possibilities illustrated in
Figure 4.
Third, despite the fact that DEPTH ASSIGNMENT and PERSPECTIVIZATION
may produce the same effects (cf. the interpretations of tief(er) and vorhinter in (9 c)-(9 d)), their distinctness on the semantic level is linguistically
overt in that, e, g., tief embodies the former while vor-hinter embody the
latter.
Fourth, on the conceptual level, PERSPECTIVIZATION involves two assignment strategies, i.e., ENCOUNTER and COINCIDENCE SITUATION, which
are to be considered as cognitively distinct devices. However, the fact
that the interpretations of vor and hinter involve the availability of either
strategy is not a sufficient reason for stipulating correspondingly distinct
semantic representations of vor and hinter. In contrast with the case of
dimension vs. distances adjectives discussed above, there is no linguistic
evidence available that would force us to assume separate lexical entries
to represent, say, an "encounter-vor" and a "coincidence-vor", and the
same with hinter. In short, as regards the semantics of vor-hinter, I claim
that the assignment strategies at issue do not figure as structural parameters on the level of semantic representation but are part of the procedures
by means of which semantic parameters are processed on the languageindependent level of conceptual representation.
Fifth, this leaves us with two things to be represented in the Semantic
Form of the prepositions vor-hinter. (a) establishing the VOR*-region
and the HINTER*-region of a given reference object requires the object
to be PERSPECTIVIZED on the basis of the Observer Axis; (b) the PERSPECTIVIZATION has to be spelled out in terms of assigning the reference object
a front and a rear.
Thus, in analogy to the representations in (7a)-(7c) and (a)-(8b), we
may assume the following lexical entries:
[ + N, +V]; A c A x [[QUANT OBS ] 2 [ + c]]6
[ + N, +V]; A c A x [[QUANT OBS [DIST y x]] 2
[v + c]]
c. /vor/;
[ - N , - V ] ; A y A x [Loc c [Loc [0BS*0 y]]]
/hinter/; [ - N , - V ] ; A y A x [Loc ^ [Loc [0BS*0' y]]]
(10)(i) a. /tief/;
b. /tief/;
266
Ewald Lang
(10)(ii)a. vor:
b. hinter:
A A
y x (LOC (x, OBS*C (y)) ...)

y A x (LOC (x, 0 B S * 0 ' (y)) ...)
OBS is a parameter identifying an axis by coincidence with the Observer

Axis, DIST accounts for distances in the way outlined in Section 2.1.2.,
the parameters 0 B S * 0 and 0 B S * 0 ' are taken to represent sets of conditions for identifying the respective front-based and rear-based regions
of the reference object y. In spelling out these parameters we will focus
on the question: Where do fronts and backs of objects come from?
2.3. The Horizontal

For reasons discussed elsewhere (cf. Lang 1987, 1989), the third axis of
PPS is not an axis we are equipped to identify by primary perceptual
information, but is derived from, hence dependent on, the two others to
fill the gap determined by the properties of the latter. Based on this, the
Horizontal does not have any bounding points or directions. Geometrically, it is reduced to the attributes of a simple line. This is the source of
the well-known difficulties in distinguishing "left" from "right". When
an object axis is assigned horizontality, its endpoints and E' are left
unspecified with respect to instantiations of "left" or "right". All that
can be said about side assignment is that and E' may not be assigned
the same value. As a consequence, I do not have much to say about the
semantics of the projective prepositionals rechts-links {von).
The distribution of "left" and "right" on the sides of an object, however,
is (partially) dependent on the side assignments the object receives due
to its verticality features and/or its intrinsic or observer-determined
"front"-"back" features (which clearly reflects the dependency mentioned
above).
Filling in the pertinent items, we get the pattern shown in (11) and
(12). Spatial expressions that refer to the Horizontal all share a feature
of inherent relativity, as reflected, inter alia, by the additional parameter
d' induced by the parameter ACROSS in (11a) for details, see Lang
(1987, 1989); by the absence of any specificatory component in ( l i b )
such that weit refers to "distance pure" (in contrast with hoch and tief in
(7 b) and (9 b)); and by making rechts-links in (11 c) inherit the type of
dependency we observe in the meaning of breit.
267
+V]; A c A x [[QUANT ACROSS d' x] 2 [v + c]]

+V]; A c A x [[QUANT ACROSS d' x] c [ v - c]]
+V]; A c A x [[QUANT [DIST y x] 5 [v + c]]
- V ] ; A y A x [Loc [Loc [ACROSS*R y]]]
- V ] ; A y A x [Loc c [Loc [ACROSS*L y ]]]
(11)
a. /breit/ [ + N,
/schmal/[ + N,
b. /weit/
[ + N,
c. /rechts/ [ - N ,
/links/ [ - N ,
(12)
a. rechts A y A x (LOC (x, ACROSS'*R (y)) ...)

b. links A y A x (LOC (x, ACROSS*L (y)) ...)
While "R" and "L" in themselves might be assessed to be mere shibboleths, the structure of the representational framework in which they
appear makes them nevertheless informative about the place they occupy
within the system of spatial expressions.
Having thus narrowed down the basic set of semantic parameters that
are specific to the projective prepositions in (1) and (2), the next step is
to examine how these parameters receive the appropriate values in order
to cover the wide range of interpretations they can assume. We will do
this by taking up the second thread which links this approach to prepositions with the study of spatial dimension terms.
3.0. Polysemy problems

The second point in favour of extending the analysis of dimension terms
to the analysis of projective prepositions is just this: the problems of
polysemy raised by projective prepositions like ber, unter, vor, hinter etc.
are strikingly similar to the ones we face regarding the range of interpretations assignable to high, deep, wide.
Thus we would expect the representational means that account, e.g.,
for the difference between primary designation (high building, tall tower,
deep hole, *high/*talll*deep ball) and contextual specification (deep plot
of ground, tall pole) also to be appropriate to cope with the difference
between intrinsic vs. extrinsic uses of projective prepositions (cf. the poster
above/in front of the building vs. the ball is above/in front of the pole). In
short, spatial prepositions make a rewarding test case of the framework
that proved to be operative for dimension and distance adjectives and
other spatial vocabulary.
As regards the treatment of polysemy, the two-level approach to the
semantics of prepositions I propose in this paper differs markedly from
the approach taken by Herkovits (or by related work within a prototypesemantics framework, e. g., Vandeloise 1984; Langacker 1986; Hottenroth
268
Ewald Lang
1988; Lakoff 1987, 1988). A discussion in full detail would take us back
to the very fundamentals of semantics. It will suffice to mention some of
the major assumptions based on which my approach differs from the
studies just mentioned.
3.1. Modularity
While other approaches are implicit in this respect, the two-level approach
explicitly ascribes to the view that our cognitive behaviour is essentially
based on structures and processes which are determined by the interaction
of relatively autonomous systems and subsystems called "modules". This
is the basic assumption which the two-level approach attempts to elaborate in tackling crucial problems of lexical semantics like the following:
(13)
a. How to interrelate the lexical meanings of spatial terms with

the morpho-syntactic category features assigned to them?
b. How to draw a justified distinction between the linguistically
coded semantic contents of a lexical item and the range of
contextual specifications it may undergo?
c. How to distinguish between, but at the same time allow for the
interaction of, word meanings as part of linguistic knowledge
and concepts as part of everyday or world knowledge?
Evidently, these questions (touched on in 1.1. and 2.1.-2.3. above) are

aimed at specifying appropriate interfaces between different levels of
mental representation, and it is along these lines that the two-level
approach I am advocating has suggestions to give and solutions to offer.
This takes us back to the problem of polysemy.
The problem raised by the theoretical need to assume abstract "ideal
meanings" of spatial prepositions on the one hand, and by the empirical
fact that we have to cope with complex contextual factors determining a
wide range of actual interpretations of these prepositions on the other
hand, is fully acknowledged but given a different status in the theory I
propose.
The two-level approach accounts of various forms of polysemy by
making it the result of the division of labour or interaction between two
levels of knowledge representation that are differentiated from, and
specifically mapped onto each other in the process of semantic interpretation. The characteristics given in 3.2. and 3.3. below may serve as
"wanted" posters for the representation levels at issue (for details see
Lang 1989, 1990 a, b).
269
3.2. Semantic form

(SF-level) = level representing language-bound word meanings as grammatically determined sets of conditions for the identification of conceptual
elements. The SF of a lexical item represents the internal structure of its
lexical meaning as part of the item's entry in the lexicon and thus specifies
what the given item can contribute to the compositional formation of
complex structures. In fact, what was proposed in (7)(12) above are SFrepresentations of adjectives and prepositions.
Let me briefly repeat: the SF of a lexical item is composed of categorized
constants and variables representing grammatically coded parameters the
values of which are available on the level of conceptual structure. As to
their form, the SF-parameters are determined by compositionality conditions; as to substance, they have to be justified with respect to the
domain-specific values they take on the conceptual level. In the case of
dimensional designation this means that SF-level representations of object
names like tower, river, pole etc. serve as addresses for access to (generic)
object concepts to be represented on the level of conceptual structure
see (16), (21) and (23) in 4.4. below.
The SF-representations of the spatial terms proposed in (7)(12) are
composed of semantic primes that are construed to represent sets of
conditions to identify an object extension (VERT, OBS, ACROSS), an
axis-based distance between objects (VERT, OBS combined with DIST),
or an object side-based neighbourhood region (VERT*F, 0 B S * 0 etc.).
In technical terms, these components are parameters on the SF-level
which are assigned values on the CS-level.
3.3. Conceptual Structure

(CS-level) = level representing language-independent, intermodally accessible elements and complexes of conceptual knowledge. CS-level is
where (generic) concepts stored in long-term memory are processed to
create contextually specified reference instances.
On CS-level, object concepts appear in the form of OBJECT SCHEMATA
(OS), which represent the defining properties of a class of spatial objects.
OS are thus representational units of the CS-level serving as location
frames within which the parameters of the SF-level are instantiated, i. e.,
are assigned Dimension Assignment Values (DAV) or, in the case of
projective prepositions, Side Assigment Values (SAV). We will come back
270
Ewald Lang
to this in 4.5. below. Given this, various phenomena subsumed under

"polysemy" can be accounted for by what I have called the division of
labour between SF-level and CS-level. The claim is that each dimensional
term has exactly one representation on SF-level, but a range of admissible
interpretations on CS-level. This range of interpretations is provided by
exactly that subset of object schemata which provide admissible values
for the SF-representation at issue. Take for instance the putative polysemy
of tief in view of the range of interpretations it is assigned in tiefer Flu,
tiefe Nische, tiefe Wunde, tiefer Topf 'deep river'; 'deep niche'; 'deep
wound'; 'deep pan'. Within the two-level approach, there is one SFrepresentation for tief containing the relevant parameter OBS (cf. (9 a),
but there are four distinct object schemata (for river, niche, wound, pan
respectively, cf. Figure 6) which determine four different ways of assigning
the SF-parameter OBS a value on CS-level. Likewise, it is due to the
differences in the object schemata that vor in, e.g., vor dem Flulder
Nische/der Wunde/dem Topf is assigned the range of distinct interpretations it covers.
3.4. Contextual determination

In the approach proposed here, the level of Semantic Form is considered
to be the INTERFACE between grammatically determined linguistic knowledge and conceptually determined everyday or encyclopedic knowledge
about the world, while Conceptual Structure is taken to provide a unified
level of representation to which visual, auditory, motor, and, of course,
linguistic information can be related and hence be interrelated among
each other. Instead of re-telling the whole story of contextually determined
dimensional designation (cf. Lang 1989, 1990 a, b), I will just quote my
paragon example of how object schemata on CS-level deal with contextual
specification.
Note that the interpretation of a sentence like Das Brett ist breit genug
'the board is wide enough' is highly context-dependent regarding the
object extent to be identified by breit 'wide'. Now, this sentence has the
same range of interpretations when applied to the spatial situations shown
in (14)(IIII), which form non-verbal contexts, as when embedded in the
verbal contexts ... und lang or und hoch or und tief genug, aber zu dnn
provided in (15)(i-iii). Thus, either of the contextual sources produces
the same effects on CS-level, that is, (14) and (15) result in the same set
of specified object schemata (16 b) for Das Brett ist breit. Note that the
271
generic concept "Brett" 'board' is represented by the unspecified basic

schema in (16 a), which may undergo the contextual specifications shown
in (16 b). For details of OS see 4.4. below. Technically speaking, nonverbal contexts like (14) and verbal contexts like (15) are equivalent as
regards their role in determining the way in which the SF-parameter
ACROSS of breit cf. (11a) is assigned its appropriate value across
in the specified object schemata in (16 b).
(14)
Das Brett ist breit genug [the board is wide enough]
7.
b
(15)
(16)
breit = a
breit = b
breit = b
lang = b
hoch = a
tief
=a
(i) Das Brett ist breit und lang genug, aber zu dnn, (breit = a as
in I)
(ii) Das Brett ist breit und hoch genug, aber zu dnn, (breit = b
as in II)
(iii) Das Brett ist breit und tief genug, aber zu dnn, (breit = b as
in III)
'The board is wide and long/high/deep enough but too thin'.
a. "Brett" <a
max
b. (I/i)
<a
max
max
c)
sub
(unspecified basic schema)
c>
sub
(III/iii)
<a
max
sub
across
(Il/ii)
across
c>
sub
<a
max
sub
across
vert
c>
sub
obs
sub
The possible equivalence of verbal and non-verbal contexts for determining the interpretation of linguistic expressions such as breit 'wide' provides
strong evidence for the claim that the contextual restrictions involved
272
Ewald Lang
must be accounted for and checked against each other on the intermodally
accessible level of conceptual structure representation (CS-level).
As concerns projective prepositions, it is on the level of CS-representations of objects where intrinsic and deictic uses have to be accounted
for, more precisely, it is the object schema of the reference object y that
provides the relevant information determining whether or not expressions
like ist vor/hinter y [ is in front of/behind y] are assigned an intrinsic
or a deictic reading. We will come back to this in section 4.4. and 4.5.
below.
3.5. Interim balance

The two-level approach expounded so far leads to a model of semantic
interpretation that comparises three major components:
(18)
a. an inventory of semantic parameters of SF-level

(including Dimension Assignment Parameters (DAP) and Side
Assignment Parameters (SAP))
b. an inventory of Object Schemata on CS-level providing sets of
values for these SF-parameters
c. a suitable device for mapping the former onto the latter
At first sight, the model may look somewhat simple in view of the
complexity of the phenomena it is devised to account for. However, the
complexity is not abstracted away but anchored in the internal make-up
of the components (a)-(c). The apparent simplicity of the model is not
the starting point but the final result of detailed investigations into object
categorization (as summarized in Figure 5 and Figure 6 below).
Having given an idea of what sort of elements the semantic parameters
are taken to be, we will now turn to the internal structure of object
concepts and take a look at the inventory of the Object Schemata.
4.0. Object concepts

Knowledge of an object embodies knowledge of the object's spatial
dimensions, that is, the gradable characteristics of its typical, possible or
actual, extension and position in space.
As regards the role of the Vertical and the Observer Axis in this respect,
we will say an object y is given an ORIENTATION if one of of the object's
273
axes coincides with the Vertical, and an object is given a PERSPECTIVIZAif one of the object's axes is identified by being aligned to the
Observer Axis.
TION
4.1. Orientation
There are various ways in which objects make reference to the surrounding space via ORIENTATION. Thus the distribution of hoch subdivides 3D
objects into four subclasses of objects with respect to having:
(19)
a.
b.
c.
or d.
fixed orientation
canonical orientation
inherent orientation
being unspecified as to orientation
(mountain, river)
(tower, desk)
(book, picture)
(brick, pole)
Let me briefly comment on these properties by focussing on the interrelations between dimension, position, movability, and side assignment.
An object is said to have a fixed orientation if has an axis which
shares all of the properties of the Vertical, specifically, if the foot F of
the Vertical located on the earth's surface is instantiated as the "bottom"
of o, without any shift or projection. This, of course, restricts the class
of objects which may be an element of to those objects which in a way
form material shells or embodiments of the Vertical per se. Therefore,
the objects that have a fixed orientation are designated parts of the earth's
surface, folded upwards, which are either natural (mountain, hill, dune)
or man-made (rampart, embankment). Being conceived of as material
embodiments of the Vertical implies that these objects do have both a
fixed position and a fixed location within Primary Perceptual Space, i.e.,
they cannot be moved or displaced or tilted etc. In short: fixed orientation
correlates with immobility. Both are non-detachable features of the conceptual representation of the objects concerned.
An object is said have a canonical orientation if has an axis which
is assigned verticality on the basis of transferring the properties of the
Vertical listed in 2.1.1. to an axis of the object at issue. The mapping is
defined by a function C which takes the end-points , E' of the object
axis as arguments and yields the specified end-points and as values,
where and are interpreted as fixed projections of the points F and
F' of the Vertical. Thus, F projected onto yields B, which is instantiated
by "bottom"; while E' (which, due to the transfer of the geofugal direction
of the Vertical, is a projection of F') is specified as yielding "top".
274
Ewald Lang
Verticality, thus assigned, implies that the objects concerned are simultaneously assigned a normal position which is defined in reference to the
Vertical of the surrounding space. Objects having a canonical orientation
may thus of course be moved, turned around, tilted etc. but any position
of theirs not meeting the conditions of fixed projection has to be judged
as marked (e.g., a desk standing upside down) or even deviant (e.g., a
tower lying after having toppled down, as suggested by long tower). In
short, having canonical orientation is part of the conceptual representation of the objects concerned.
An object is said to have an inherent orientation if has an axis
which is assigned verticality in a way similar to the one described above,
but with the crucial difference that "top" and "bottom" are instantiated
in the object's own right, that is, that the inherent height of object no
longer has any links with the Vertical of the surrounding space. Thus a
book, for instance, has an inherent height (and therefore an inherent
"top", "bottom", "beginning", "end") due to the inscription it carries.
Likewise, a photograph showing the Queen or a button showing Gorbi
have an inherent verticality which is maintained absolutely independently
of the object's actual position (standing, lying, hanging upside down etc.).
Inherent verticality of an object is thus to be conceived as a way of
orientation which, though originally derived from the Vertical, has become independent of the axes of Primary Perceptual Space due to pragmatic conditions. This explains why a difference between the intrinsic
and actual verticality of an object o, say, if is tipped over, can produce
quite distinct results as to conceptualization. To imagine a mountain as
tipped over or standing on edge is almost impossible; to conceive a tower
as long (i.e., as lying down) involves loosening the concept "tower";
whereas to see a book lying on the table or standing upside down does
not at all affect our understanding of "book". At best, the difference
between the inherent and the actual verticality of a book may be taken
as an indicator of pragmatic deviation.
An object is said to have a contextually induced orientation if has
an axis which is assigned verticality in such a way that the projection of
F and F' onto the endpoints and E' of an axis of o, and therefore the
instantiation of "bottom" and "top", apply to the actual position of the
object at issue. Verticality can be induced contextually with objects that
in themselves are unspecified as to orientation (cf. a long pole specified
as a tall pole, a brick specified as being 17 cm in height etc.) or that have
an inherent orientation. The latter case causes the multiple interpretation
of, e.g., The tombstone is too high to fit into the trunk, where high can
275
either refer to the tombstone's inherent top-bottom extent or to its

contextually specified thickness. With objects having a canonical orientation, contextually induced verticality makes sense only if the object at
issue is not in normal position. Otherwise, the parameter VERT (cf. tall
tower, low table) will always be instantiated by the canonically oriented
object axis.
Contextually induced orientation thus amounts to specifying the actual
position of an object with reference to the Vertical. Normally, this applies
to freely movable objects which do not have an intrinsic (fixed or canonical) orientation. At the conceptual level, the difference between intrinsic
and contextually induced orientation reappears in the way the parameter
VERT is instantiated at CS-level: in the first case an intrinsic verticality
feature of an object axis is identified, in the second case a suitable object
axis is specified by inducing a verticality feature. The instantiation of a
SF-parameter by either identification or specification of a value on CSlevel is also of central import to the analysis of projective prepositions
cf. 4.5. below.
4.2. Perspectivization
Moving on to PERSPECTIVIZATION we find a very similar picture, except
that there seems to be one type of assignment missing with OBS. The
distribution of tief subdivides 3D objects into three subclasses of objects
with respect to having:
(20)
a. canonical perspectivization
b. inherent perspectivization
or c. being unspecified as to perspective
(river, cupboard)
(hole, wound)
(brick, board)
The first point to note is that here there is no analogue to fixed orientation
in (19 a). Unlike the Vertical, the Observer Axis is not constant and
independent, but flexible in two respects. It has a pivot allowing for a
180 turn in the vertical and the horizontal plane, and it has a moving
source which taking upright posture as normal position is at least
partially determined by its relation to the Vertical. This has the consequence that there is no way of assigning an object a fixed perspective in
exactly the same sense as an object may have a fixed orientation. It takes
but a moment's reflection to see why.
In view of the flexibility of the Observer Axis, it holds that the position
it assigns to an object is also determined by the position of the observer.
276
Ewald Lang
Hence there cannot be any object which embodies depth in such a way
that this alone would suffice to locate in the surrounding space and
simultaneously determine its position. Note the rule: Perspectivization is
always linked with orientation. So much for why the analogue of FIXED
ORIENTATION is absent with DEPTH ASSIGNMENT.
An object is said to have a canonical perspectivization with respect
to the Observer if o, on account of its intrinsic orientation according to
(19 a) or (19 b), also has an axis that is designated as being aligned to
the Observer's line of sight. Based on the fact that there are two relevant
angles at which the Observer Axis may run to the Vertical, we have to
distinguish two cases of canonical perspectivization.
The first case is exemplified by objects such as cupboard or doorway
(cf. Figure 6 below), where the depth axis is orthogonal to the canonically
assigned vertical axis. These objects are movable but have a canonical
position determined by their orientation and, depending on that, by their
perspectivization. The mapping of O' and O" onto the endpoints and
E' of the axis concerned follows the ENCOUNTER strategy (shown in Figure
4) in such a way that O' is mapped onto and instantiated as canonical
"front"; O" is mapped onto E' and instantiated as canonical "back".
These are sides which may be accessed by the parameters 0 B S * 0 and
0 B S * 0 ' contained in vor and hinter.
The second case is exemplified by objects such as river, ditch or well
(cf. Figure 6 below), where the depth axis runs at an angle of 180 to the
Vertical, that is, the object axis designated as depth axis continues the
Vertical from F in the opposite direction ("downward"). The function D
that maps the bounding point of the Observer Axis onto the foot F of
the Vertical and takes account of the opposite directions in an appropriate
way is discussed at length in Lang (1987, 1989: Chapter 2). Note that
the immobility of objects such as river or ditch is due to their having a
fixed orientation in the sense of (19 a), that is, a fixed relation to the
Vertical similar to objects such as mountain or rampart. Thus, mountain
and river group together in being designated parts of the earth's surface,
and therefore being immobile objects, but they differ in their relation to
the Observer Axis.
The fact that the depth axis assigned to a river continues the Vertical
in the opposite direction does have its consequences regarding the mapping of and O' onto the end-points of the object axis at issue. In short:
the mapping also follows the ENCOUNTER strategy described above, but
with the additional effect (induced by the Vertical) that O, when mapped
onto E, is instantiated as "accessible surface" (which conflates "top" and
277
"front"); while O', when mapped onto E', is instantiated as "ground"

(which conflates "bottom" and "remote face"). In other words, this
explains why a river or a ditch cannot be said to have front and rear
sides or top and bottom sides.
This view of CANONICAL PERSPECTIVIZATION seems to match with the
way depth is conceptualized, not least due to the fact that it reflects the
dominant role of the Vertical and the subsidiary role of the Observer
Axis.
An object is said to have an inherent perspectivization if y has a
disintegrated axis which can only be identified by being aligned to the
Observer Axis. In other words, if there are no other cues available for
assigning a position property (by relating the axis at issue to the Vertical)
or for assigning a maximal or substance-determined axis, then precisely
this object axis is designated as depth.
Note that, in line with INHERENT ORIENTATION in (19 b), the notion of
INHERENT PERSPECTIVIZATION is basically defined as "assigning an axis of
object the properties of an axis of PPS without regard to the object's
actual position in space". Regarding inherent depth, this is exactly the
case with either immobile hollow objects, e. g., holes, or freely movable
objects with an interior, e.g., cups, pots, bottles etc. A (non-perforated)
hole, no matter whether it is in the floor, in the wall, or in the ceiling, is
invariably assigned depth. As the example suggests, a hole or the interior
of a cup is conceived as a designated (negative or missing) part of the
solid object containing it. Therefore, a hole or the interior of a cup cannot
move or change position by itself. Given this, it is due to the flexibility
of the Observer Axis that such objects can be inherently assigned depth
without regard to their location in the surrounding space.
The mapping of and O' onto and E' proceeds along the lines of
the ENCOUNTER strategy shown in Figure 4 above, that is, is mapped
onto and instantiated as "front" or "accessible surface"; while O' is
mapped onto E' and instantiated as "rear" or "remote face".
An object is said to have a contextually induced perspectivization if
has an axis which is assigned depth due only to the actual position of
the object at issue, that is, if neither (20 a) nor (20 b) apply. As was
discussed above, there are two ways of mapping and O' onto the endpoints of the object axis that is designated as depth axis. This claim
would predict that contextual perspectivization, when applied to objects
that in themselves are unspecified as to perspectivization, should yield
two different assignments of "front" (or "accessible face") and "back"
(or "remote face"), resulting from either the ENCOUNTER SITUATION or the
278
Ewald Lang
And this is indeed the case cf. the discussion

in Section 2.2. above.
The different ways of perspectivizing an object discussed so far are
closely related to the familiar distinction of INTRINSIC VS. DEICTIC assignment of sides to objects. As can be easily seen, objects that have a
CANONICAL or INHERENT PERSPECTIVIZATION along the lines of (20 a) and
(20 b), have, in consequence, INTRINSICALLY determined "fronts" and
"backs" (or, "accessible surfaces" and "remote faces"), construed as
instantiations of the end-points of that object axis that is designated as
being aligned to the Observer Axis. Given this, it is up to CONTEXTUALLY
INDUCED PERSPECTIVIZATION to account for DEICTICALLY determined
"fronts" and "backs".
COINCIDENCE SITUATION.
4.3. Object categorization

We summarize the facts about ORIENTATION and PERSPECTIVIZATION and
their interrelations to an object's (im-)mobility by the following diagram
shown in Figure 5. To give an indication of what it really means, some
brief comments seem to be in order.
First, for the purpose of transparency, the division of objects into
immobile objects and movable objects is indicated by grey and white
backgrounding, respectively. Notice, however, that these mobility characteristics are not independent, but result from the values each object
assumes with respect to boundedness, dimensionality, and relation to
PPS-axes.
Second, the terminal nodes of the taxonomy, thus determined, are
exemplified by illustrative samples of object concepts. They give an idea
of what information is included in the Object Schemata (cf. Section 4.4.).
Third, the categorization of objects in Figure 5 is designed to account
for the range of conditions according to which the parameters VERT
and OBS are instantiated by values at the CS-level. The division of spatial
objects into four classes, that is, immobile objects on the one hand, and
three classes of movable objects on the other, directly reflects the range
of positional variation each of these classes of objects can be subjected
to. Put in terms of, say, lying, standing, tilted etc., the following holds:
immobile objects do not allow for any variation or change of their
position; canonically oriented objects allow for certain changes in their
verticality assignment (a tower can be said to lie after having toppled
down; roots can be said to lie when they have been dug of the ground;
279
a shelf can stand upside down etc.); inherently oriented objects allow for
all changes of position {lying, standing, upside down), and so do contextually oriented and/or perspectivized objects.
Fourth, the parameter ACROSS, as will be remembered, is in any case
assigned a contextually induced value (cf. the remarks in 2.3.). Due to
the fact, however, that ACROSS is a stop-gap, it is neutral with respect
to mobility features and hence applies across movable and immobile
objects (as indicated by grey backgrounding in Figure 5).
Fifth, Figure 5 offers a complete categorization of objects with respect
to intrinsically or/and deictically assigned sides. It is this information that
is accessed by the parameters VERT*F', VERT*F in ber-unter and/or
by 0BS*0, 0 B S * 0 ' in vor-hinter. The overlapping area marked by a
grid pattern at the bottom of Figure 5 shows the range of objects that
are available for both intrinsic and deictic assignment of sides.
4.4. Object Schemata

The most plausible way in which conceptual representations of spatial
objects may be construed is to render them in the form of a matrix with
3 rows and 1, 2 or 3 columns (depending on the nature and number of
the axes of the object represented). We call such a matrix an Object
Schema (OS). An OS contains entries which represent the defining properties of a class of spatial objects. OS are thus representational units of
the CS level serving as location frames within which the parameters of
the SF level are instantiated. The structure of an object schema and the
selection of entries it may contain are again entirely determined by the
categorization shown in Figure 5. Let me exemplify the make-up of OS
by commenting on the OS that represents the object concept "cupboard":
(21)
cupboard
<a
vert
F', i-top
F, i-bottom
b
across
S, i-les
S', i-ris
c)
obs
O', i-back
O, i-front
Take the head row first. In containing one, two, or three variables for
object axes (a, a b, a b c) it represents the dimensionality of the object
at issue, the boundedness by angled brackets, the integration of axes by
brackets. This yields exactly 7 types of OS as shown in Figure 6 below.
280
Ewald Lang
The second row reflects defining gestalt and position properties of the
given object, i.e., entries such as vert, i-vert, obs, i-obs, max, sub etc.
Being part of an OS, these symbols now stand for conceptual values
instantiating the parameters VERT, OBS etc. of the SF-level (note the
typographic distinction). It is these entries which account for the different
Spatial objects
movable
Boundedness
-immobile-
unbounded
"air"
"water"
- i "gold"
bounded
Dimensionality
J
3D
2D
"plot"
"field"
0Relation to PPS -Axes9

contextual])'
induced
OS entry
"line"
"edge"
intrinsic
inherent
canonical
orientation orientation
or/and perspectivization
(
ID
J i-
obs v e r t
-m
vert
obs
fixed
orientation
inherent
perspectivization
i-obs
~H
Intrltt&fcVtty assigned sijles or/qn'd regions

access side
remote side
orientation and/or
perspectiv ization
d e i c t i c oily assigned
sides and/or regions
Figure 5. The interaction of dimension, position, and side-based region asignment to spatial
objects
281
ways according to which an object is ORIENTED or/and PERSPECTIVIZED

(cf. 4.1. and 4.2. above). For instance, the entry vert in (21) indicates the
object's canonical orientation, an entry i-vert inherent orientation (cf.
"book" in Figure 5), an entry f-vert fixed orientation (cf. "mountain" in
Figure 5); and similarly so for PERSPECTIVIZATION by means of entries
like obs or i-obs. For further objects the OS of which contains such entries
see Figure 5 and Figure 6.
Based on the entries, which specify the object's axes, the OS contain
information on the end-points of the pertinent object-axes (cf. F', F; S,
S'; ', in (21)) and on the intrinsic sides of the object defined by these
end-points. Thus, the canonical height-axis of the cupboard (indicated
by vert) has the end-points F', F which, in turn, determine the cupboard's
intrinsic top and bottom sides, indicated by "i-top" and "i-bottom",
respectively. Similarly, the endpoints O, O' of the canonically perspectivized depth axis of a cupboard are used to determine the object's intrinsic
front (i-front) and rear (i-back), respectively. The end-points of the third
axis, which by themselves would not have designated values due to the
specific nature of across, are in the given context of the other OSsections marked as carriers of intrinsic lefthand side (i-les) and righthand side (i-ris), respectively.
The third row (below the line) of an unspecified generic OS contains
slots with variables (omitted here for transparency's sake) to be replaced
with contextual specifications induced by linguistic or situational contexts
(Cf. the examples in (14)(16) above). In the course of processing an OS,
the third row is filled with appropriate values.
Figure 6 below (quoted from Lang et al. 1991) presents a complete
catalogue of those admissible object schemata for which we were able to
find some empirical instances. Well, we would not argue about adding
one or two further OS that might have escaped us. But if this is granted,
we may claim that the catalogue of OS presented below is exhaustive as
regards those spatial properties that are involved in dimensional designation, positional variation, and intrinsic ORIENTATION and PERSPECTIVIZATION of objects.
Now, let me briefly outline how the meaning of the projective prepositions interacts with the information contained in the OS. To begin with,
it is the entries on intrinsic sides in an OS which are accessed to by the
SF-parameters contained in the projective prepositions ber-unter, vorhinter, rechts-links. Thus the semantic interpretation of, say,
(22)
Der Tisch ist vor dem Schrank.

'The table is in front of the cupboard.'
282
Ewald Lang
Object
Schemata
Spatial Objects
I
bounded
a. b, c
dimensionality
integration of
axes
No
< (a
dist
c >
max vert sub
"yard"
"hole"
"wall"
max acros
diam
max i-vert sub
"sidewalk"
"circle"
"ruler"
max i-vert
max vert obs
"banner"
"ledge"
vert across
"door"
f-vert dist
"doorway*
0
vert
"window"
0
i-vert
"picture"
max vert across
< a
max
sub
"pole"
vert sub
"cylinder"
max
subver
"tower"
max
dist
"pipe"
max across vert

obs
max
distvert
"river"
0 vert
"banquet table"
I
VII
1
VI
c) > < (a
(b
"Great W a l l "
max
"line"
1
V
IV
b) > < a
max
Yes
No
b >
max
integration
of axes
Yes
II
< a
unbounded
"bottle"
b)
c >
diam
"disk"
"hall"
"tart"
diam f-vert
"hill"
dist sub
"hoop"
dist vert
"saucepan"
"bore hole"
"hole"
"niche"
"barrel"
vert across obs
vert dist
obs
"cupboard"
"well"
c )>
diam vert
dist i-obs
vert dist
diam sub
i-obs dist
vert across i-obi
< (a
dist vert
obs
"den"
Figure 6. Categorization of spatial objects at the conceptual level and inventory of object
schemata
IV
<a
c>
max across vert

i-obs
"mine hole"
across obs vert
"stair"
max
vert
obs
"elevator shaft"
i-vert 0
sub
"book"
0
vert
obs
"chest"
obs dist vert

"doorway"
0
sub
"brick"
0
i-obs
"wound"
0
0
"coffin"
[ Fig. 6 ' ] Further OS of type IV
vert
283
284
Ewald Lang
proceeds in the following way. Based on general rules of composition,

the SF-representation of vor, i.e., A y A x (LOC (x, 0 B S * 0 (y)) ...), is
linked to the SF-representation of the reference object denoted by
Schrank. To avoid thorny problems of delimiting linguistic knowledge, I
have not discussed the proper lexical meaning of object names like
Schrank. (It might be conceived as a set of parameters that range over
the properties encoded in the top two rows of the corresponding OS). At
any rate, the compositional amalgamation of the SF-representations of
vor and dem Schrank makes the OS associated with the latter available
to the SF-parameter 0 B S * 0 of vor such that the OS shown in (21)
provides a proper subset of the range of values that instantiate 0 B S * 0 .
In the given case, the first choice to evaluate 0 B S * 0 is to assign it the
value "i-front" from the OS in (21). The result is that (22) is given an
interpretation of vor that corresponds to the so-called intrinsic use, that
is, an interpretation that involves a VOR*-region which is determined by
the intrinsic front of the reference object. Due to the nullo obstanteprinciple, this way of evaluating the parameter 0 B S * 0 would yield the
default interpretation of (22), that is, if there is no indication that the
cupboard is not in its normal position as encoded in the top two rows
of the OS. The default evaluation results in a specified OS containing a
second entry "d-front", which is inserted in the third row, third column
under "i-front" to indicate that intrinsic front and actual front of the
given reference object are matching (cf. (23 a)).
But (22) is open also for other interpretations. Suppose a situation
where a cupboard is laid down and tilted to the back and the table is
now facing the intrinsic bottom side of the cupboard. Given these situational framework, the evaluation of the parameter 0 B S * 0 in vor would
lead to a different result. In fact, the OS of "cupboard" will notify in the
third row the contextually induced changes with respect to the entries
contained in the top two rows. What we get is a specified OS of Schrank
that shows all the differences between intrinsic and actual side assignments
of the given object cf. (23 b). For transparency, I indicate only the
relevant entries in the third row.
(23)
a. vor dem Schrank (default interpretation)

<a
vert
F', i-top
F, i-bottom
b
across
S, i-les
S \ i-ris
c>
obs
O', i-back
O, i-front
d-front
(23)
285
b. vor dem Schrank (with cupboard laid down and tilted to the
back)
<a
vert
F', i-top
F, i-bottom
b
across
S, i-les
S', i-ris
c>
obs
O', i-back
0 , i-front
d-back,
d-front
i-les,
i-ris
d-bottom,
d-top
Note that the specified OS in (23 b) contains all and only the relevant
values that determine the interpretation of (22) under the assumed situational setting. Along the same lines we obtain any other contextually
determined interpretation that (22) can assume. In other words, by
allowing for the parameter 0 B S * 0 to take both values that match with
intrinsic OS-entries (cf. (23 a)) and values that differ from, but are compatible with, intrinsic OS-entries (cf. (23 b)), we are able to account of
the full range of interpretations which vor can assume in localizing a
given object in the VOR*-region of the reference object y.
Moreover, the interpretation of hinter, ber, unter proceeds along the
same lines. The reader is invited to examine the complete list of Object
Schemata given in Figure 6 in order to check himself which objects
provide which intrinsic sides that may serve to determine object-based
neighbourhood regions. This much as to structure and inventory of OS,
which form, as claimed in (18), the second component of the two-level
approach I am advocating. We will now turn to the third component.
4.5. The mapping device
Needless to say that the evaluation of SF-parameters like VERT, OBS,

VERT*F, VERT*F', 0BS*0, 0 B S * 0 ' etc. is to be construed as a restricted mapping from elements on the SF-level onto elements of the OS
on the CS-level.
The mapping device comprises two basic operations called identification and specification of semantic parameters (DAP and SAP) with
respect to the set of values (DAV and SAV) they may assume on the CSlevel. I will refrain from re-telling the story of how it works with dimension
286
Ewald Lang
terms (cf. Lang 1987, 1989 and Lang et al. 1991), though this forms the
background, but instead move directly to the projective prepositions.
Suppose the set of semantic parameters needed to account for the
projective prepositions in (1) is given by SAP = {VERT*F, VERT*F',
0BS*0, 0BS*0', R, L}, while the set of side assignment values is given
by the set SAV = {(vert, F', i-top), (vert, F', d-top), (vert, F, i-bottom),
... <obs, O, i-front), ...} see (23). Now, an element of SAP is mapped
onto an element of SAV by way of identification iff matches the first
chunk of p. Somewhat simplified we get:
(24)
Identification: => p,
where e {VERT*F, VERT*F', 0BS*0, 0BS*0'},
e {(vert, F, i-top), (obs, O, i-front) ...} and
{<i-vert, ...), <i-obs, . . .
In contrast, an element of SAP is mapped onto an element q of SAV

by way of specification iff attaches to q in an OS-section. Attachment
is restricted by suitable conditions determining possible specifications (see
below). Somewhat simplified we get:
(25)
Specification: => q,
where e {VERT*F, VERT*F', 0BS*0, 0BS*0'},
q e {(max, _, _>, <0, _>, ...} and
q {< vert, F, i-top), (obs, O, i-front), ...} and
q is licensed as a landing site for in OS
Two remarks on the conditions constraining the mapping might be in

order. The second condition in (25) excludes OS-entries for fixed and
canonic orientation from being specified vacuously with identical or
contradictorily with incompatible values. The third condition refers to
general compatibility conditions among spatial parameters for details
see Lang Carstensen (1990). The second condition in (24), however,
prevents inherent verticality or perspectivization values from automatically instantiating VERT*F or 0BS*0. This accounts for the fact that,
e.g. inherent top or front sides (as with books) do not define proper
subregions of the reference object of a projective preposition. This can
be illustrated by the following example:
(26)
Das Dreieck ist ber dem Buch, der Ball ist vor dem Buch
'The triangle is above the book, the ball is in front of the book'
287
(a) configurations that satisfy (26)
>
>
>
t>
-)
GB
80
(b) configurations that do not satisfy (26)
>
Moreover, the restricted range of interpretations (26) may assume is

indicative of the following rule: if intrinsic and extrinsic verticality assignments are at variance, the extrinsic way of assigning top and bottom
prevails. This rule, in turn, is but another piece of evidence for the claim
that the Vertical is the dominant axis. There is an order of preference
also among the various ways of assigning objects fronts and rears but
the discussion of this issue should better be reserved for my next paper
on projective prepositions.
"The purpose of science", says Paul Valery, "is to give us a completely
different idea of the things we know best." The object-centered and axisbased approach to prepositions outlined in this paper seems to serve this
purpose almost literally.
Notes
* This is the extended version of a paper delivered at the Saarbrcken EUROTRAConference 1990. Most of the data used here are also discussed in Lang (1991), which
includes a more thorough analysis of the prepositions vor-hinter, links-rechts than in
the present study. I wish to thank C. Zelinsky-Wibbelt for her patience in editorial
matters, and several English native speakers for checking the data. Needless to say that
I am responsible for all remaining errors and uncertainties.
288
Ewald Lang
1. The term is due to Herskovits (1986: 156 ff.), who calls "these prepositions 'projective',
because all fundamentally involve the experience of viewing and the idea of a point of
observation." T h o u g h I do not fully agree with this definition, I shall stick to the term
as it has become a familiar label in the literature.
2. In comparison to German, the inventory of projective prepositions in English is (a)
richer and (b) patterns across simple and complex expressions. Thus, the equivalents of
(1) include at least the following items (for completion see Lindkvist (1972), Herskovits
(1986), Boers (1987)):
(i) above/below, over/under, on top o f . . .
in frontjat the back of, before/behind ...
right/left of, at j on/in/to/by the right/left (hand) side o f . . .
3.
4.
5.
6.
T h o u g h I will not delve into details, two points should be noted. First, in explicitly
making reference to designated sides (or parts) of objects, the complex expressions in
(i) reveal what the semantics of projective prepositions is about it is a device to
locate an object in relation to a reference object by specifying a side-based neighbourhood-region of the latter. Second, there is a wide range of uses where simple and
complex projective expressions are interchangeable, but there are also differences, e. g.,
the puzzling fact that in front of, on top of etc. hardly combine with measure phrases,
whereas the G e r m a n adverbials in (2) do.
But note the following asymmetry in (1). While for (a) and (b), as with other prepositions,
local vs. directional uses correlate with dative vs. accusative, the pair in (c) is different
in requiring genitive for local use and in lacking directional use altogether. I take this
as evidence in support of the claim that the three axes involved (viz. Vertical, Observer
Axis, Horizontal) differ significantly as to their role in the conceptualization of space
and hence as to the way they figure in the semantic structure of spatial expressions (see
Sections 2-A).
Based on evidence from, e. g., French or Spanish, where the translation of (6 a) would
yield two morphologically distinct constructions, I assume two separate lexical entries
for hoch (or high) corresponding to dimension and distance interpretation, respectively.
Cf. (7 a) and (7 b).
I refrain f r o m discussing the details with respect to which German here differs f r o m
English or French.
The dimension adjective tief lacks a lexical antonym, for discussion cf. Lang (1987,
1989).
References
Bierwisch, Manfred
1988
"On the g r a m m a r of local prepositions", in: M a n f r e d Bierwisch W. Mtsch
I. Z i m m e r m a n n (eds.), Syntax, Semantik und Lexikon. 1-65. (studia grammatica 29) Berlin: Akademie-Verlag.
1987
Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven,
(studia
grammatica 26 + 27) Berlin: Akademie-Verlag.

1989
289
Dimensional adjectives: Grammatical structure and conceptual interpretation.

(Springer Series in Language and Communication 26) Berlin-HeidelbergNew York: Springer-Verlag.
Boers, Frank
1987
Behind, beyond, under, underneath, beneath, below: A descriptive and explanatory study of spatial and non-spatial senses. Antwerp Papers in Linguistics
53. University of Antwerp.
Carstensen, Kai-Uwe
1991
Aspekte der Generierung von Wegbeschreibungen. (IWBS Report 190) Stuttgart: IBM Deutschland.
van Doom, Antonius J. W. A. van de Grind J. J. Koenderink (eds.)
1984
Limits in Perception. Utrecht: VNU Science Press.
Habel, Christoph
1989
"zwwc/ien-Bericht", in: Christoph Habel Michael Herweg Klaus Rehkmper (eds.), 37-69.
Habel, Christoph Simone Pribbenow
1988
Gebietskonstituierende Prozesse. (LILOG-Report 18) Stuttgart: IBM Deutschland.
Habel, Christoph Michael Herweg Klaus Rehkmper (eds.)
1989
Raumkonzepte in Verslehensprozessen. (Linguistische Arbeiten 233) Tbingen:
Niemeyer.
Haumann, Dagmar
1992
Lexikalische Argumentstruktur und -Rollen englischer
Lokalprpositionen.
Arbeiten des SFB 282 "Theorie des Lexikons" 15, Universitt Wuppertal.
Herskovits, Annette
1986
1988
"Spatial expressions and the plasticity of meaning", in: Brigida Rudzka-Ostyn
(ed.), 403-427.
Herweg, Michael
1989
"Anstze zu einer semantischen Beschreibung topologischer Prpositionen",
in: Christoph Habel Michael Herweg Klaus Rehkmper (eds.), 99-127.
1990
Zeitaspekte. Die Bedeutung von Tempus, Aspekt und temporalen Konjunktionen.
Wiesbaden: Deutscher Universitts-Verlag.
Herzog, Otthein Claus-Rainer Rollinger (eds.)
1991
Text understanding in LILOG: Integrating computational linguistics and artificial intelligence. (Lecture notes in artificial intelligence 546.) Berlin-Heidelberg-New York: Springer-Verlag.
1988
Die Semantik lokaler Prpositionen. Ein prototypensemantisches Modell fr
die franzsische Prposition dans mit einer Analyse der Beziehungen zwischen
der Prposition und den Objektbezeichnungen in den Prpositionalsyntagmen. [Habilitationsschrift, Universitt Konstanz.]
Kaufmann, Ingrid
1989
"Direktionale Prpositionen", in: Christoph Habel Michael Herweg
Klaus Rehkmper (eds.), 128-149.
1990
"Semantische und konzeptuelle Aspekte der Prposition durch", Kognitionswissenschaft 1: 15-26.
290
Ewald Lang
Klose, Gudrun Ewald Lang Thomas Pirlein (eds.)

1992
Ontologie und Axiomatik der Wissensbasis von LILOG. (Informatik-Fachberichte 307.) Berlin-Heidelberg-New York: Spinger-Verlag.
Lakoff, George
1987
Chicago: The University of Chicago Press.
1988
"Cognitive Semantics", in: U. Eco, M. Santam Crogio, P. Violi (eds.), Meaning
and Mental Representations. 119-154. Bloomington: Indiana University Press.
Lang, Ewald
1988
Object Schemata and Rumliche Konfigurationen. "Rumliche Lokalisierung"
Project. Universitt Dsseldorf.
1989
"The semantics of dimensional designation of spatial objects", in: Manfred
Bierwisch Ewald Lang (eds.), 263^17.
1990 a
"Primary perceptual space and inherent proportion schema", Journal of
Semantics 7: 121-141.
1990 b
"Sprachkenntnis, Objektwissen und rumliches Schlieen", Zeitschrift fr
Linguistik und Literaturwissenschaft 78: 59-97.
1991 a
"A two-level approach to projective prepositions", in: Gisa Rauh (ed.), 127167.
1991 b
"The LILOG-ontology from a linguistic point of view", in: Otthein Herzog
Claus-Rainer Rollinger (eds.), 464-481.
1992
"Linguistische vs. konzeptuelle Aspekte der LILOG-Ontologie. Anfragen von
auen", in: Gudrun Klose Ewald Lang Thomas Pirlein (eds.), 23-45.
Lang, Ewald Kai-Uwe Carstensen
1989
"OSKAR ein Prolog-Programm zur Modellierung der Struktur und Verarbeitung rumlichen Wissens", in: Dieter Metzing (ed.), 234-243.
1990
OSKAR A prolog program for modelling dimensional designation and positional variation of objects in space. (IWBS Report 109.) Stuttgart: IBM
Deutschland.
Lang, Ewald Kai-Uwe Carstensen Geoffrey Simmons
1991
Modelling spatial knowledge on a linguistic basis. Theory Prototype
Integration. (Lecture Notes in Artificial Intelligence 481) Berlin-HeidelbergNew York: Spinger-Verlag.
1987
Foundations of cognitive grammar. Vol. I. Theoretical prerequisites. Stanford,
CA: Stanford University Press.
Levelt, Willem J. .
1984
"Some perceptual limitations on talking about space", in: A . J . van Doom
- W. A. van de Grind - T.T. Koenderink (eds.), 323-358.
Li, Jie
1991
Rumliche Relation und Objektwissen. Untersuchung am Beispiel der
deutschen Prpositionen an und bei. [Dissertation Eberhard-Karls-Universitt
Tbingen.]
Lindkvist, Karl-Gustav
1972
The local sense of the prepositions OVER, ABOVE, and ACROSS studied in presentday English. (Stockholm Studies in English XXV) Stockholm: Almquist &
Wiksell.
291
Maienborn, Claudia
1989 a
Zur Semantik von Bewegungs- und Positionsverben. Perspektiven der kognitiven
Linguistik. (LILOG Report 64) Stuttgart: IBM Deutschland.
1989 b
Position und Bewegung: Zur Semantik lokaler Verben (IWBS-Report 138)
Stuttgart: IBM Deutschland.
1991
"Verbs of motion and position: On the optionality of the local argument",
in: Otthein Herzog Claus-Rainer Rollinger (eds.), 621-631.
Metzing, Dieter (ed.)
1989
GWAI-89. 13th German workshop on artificial intelligence. (Informatik-Fachberichte 216). Berlin-Heidelberg-New York: Springer-Verlag.
Rauh, Gisa
1991 a
"Prepositional forms in the lexicon: problems and suggestions", in: Gisa
Rauh (ed.), 169-224.
1991 b
Zur Grammatik lexikalischer und nicht-lexikalischer Prpositionen im Englischen. Arbeiten des SFB 282 "Theorie des Lexikons" 5. Universitt Wuppertal.
Rauh, Gisa (ed.)
1991
Approaches to prepositions. (Tbinger Beitrge zur Linguistik 358) Tbingen:
Gunter Narr Verlag.
Retz-Schmidt, Gudula
1988
"Various views on spatial prepositions", AI Magazine 9: 95-105.
Rudzka-Ostyn, Brigida (ed.)
1988
Topics in cognitive linguistics. Amsterdam: Benjamins.
Simmons, Geoffrey
1991
"Empirical methods for ontological engineering", in: Gudrun Klose Ewald
Lang Thomas Pirlein (eds.), 125-154.
1992
Standardwissen ber Normen: Zur konzeptuellen Analyse von Objekten. (IWBSReport 000) Stuttgart: IBM Deutschland.
von Stechow, Arnim Dieter Wunderlich (eds.)
1991
Semantik/Semantics.
Ein internationales Handbuch der zeitgenssischen Forschung. (Handbcher zur Sprach- und Kommunikationswissenschaft 6) Berlin-New York: de Gruyter.
Vandeloise, Claude
1986
L'Espace en frangais. Semantique des pripositions spatiales. Paris: Editions du
Seuil.
Wunderlich, Dieter Michael Herweg
1991
"Lokale und Direktionale", in: Arnim von Stechow Dieter Wunderlich
(eds.), 758-785.
German temporal prepositions from

an English perspective
Martin Durreil David
Bree1
The following is an attempt to analyse the procedures for selecting

temporal prepositions (or conjunctions) in German according to the
framework developed in Bree et al. (1990). Systematic contrast is made
with the English as analysed there (and, where interesting or relevant
with Dutch). The German time words follow similar selection trees for
those in English and Dutch, but there are some quite interesting differences (and not a few problems), and some amendments had to be made
to the analysis in Bree et al. (1990). Despite these differences, there would
seem to be no reason in principle why the analytical framework in terms
of decision trees adopted by Bree et al. (1990) should not be as revealing
for German as for other languages.
An initial analysis was made in terms of equivalents established in the
course of work on Durrell (1991 a and 1991 b) rather than by looking at
translation equivalents. This had the advantage that no major equivalents
were missed, and other authoritative data sources were also checked to
this end, viz. Buscha (1989), DUDEN (1976-81), Oxford-Duden (1990),
Schmitz (1964), Schrder (1990) and Terrell et al. (1991). However, this
procedure meant that assessments regarding the relative standing of the
equivalents were subjective, whilst in Bree et al. (1990) they were established objectively through a corpus of translation data. As a consequence,
there may be some skewing because of the lack of comparable corpus
data and this could have led to undue importance being given to some
equivalents. Any possible failings here would need to be rectified later
on the basis of a corpus of translation data.
1. The English and Dutch selection trees

Bree et al. (1990) proposed selection trees for English and Dutch temporals on the basis of translation data, and these we show in Figures 1
and 2. We now describe the branches of the English and Dutch selection
trees.
296
Martin Durreil David Bree
M,S
M a i n Stevern t i m e
D u r a t i o n of M , S
S u b stevent t i m e
TOD
T i m e of D i s c o u r s e
Open choice
Figure 1. Selection tree for English temporal function words (revised)
1.1. The selection tree for English temporal function words

The first and main distinction is whether or not the time of the matrix
state or event, "stevent", is being equated to that of the sub stevent. If
it is, then the nature of the sub stevent is considered. If the sub stevent
is:
an event then:
use as;
unless this event shifts the Time of Discourse, TOD, then:
if the sub event is a period, use while,
otherwise, use when;
a state, then:
if it is a time point, then use at,
unless the matrix stevent can take place at a time earlier than the
sub state, then use by;
German temporal prepositions

S is a
period
single
in past
^
297
}/lang
voor (diH)
daarvoor/voordien
M a i n slcvont time
S u b Stevern lime
TOD
T i m e of D i s c o u r s e
M.S
Duration of , S
Figure 2. Selection tree for Dutch temporal function words (revised)
if the sub state is selected from a series of states, e. g., days of the
week, then use on;
if the sub state has a duration then:
if the duration of the matrix stevent is to be equal to that of the
substate, then use for;
if the duration of the matrix stevent is to be less than that of the
sub state, then use within;
otherwise, use in,
unless the sub state can also be given a non-temporal interpretation (usually locative), then use during.
If the time of the matrix stevent is not to be made approximately the
same as that of the sub stevent, then the order of the matrix and sub
stevents needs to be considered. If the matrix stevent is to fall in a period
that:
is to follow the sub stevent, then:
if the other end of the period is to be the TOD, then use since
otherwise use after;
298
Martin Durrell - David Bree
is to precede the sub stevent, then

if the other end of this period is to be the TOD, then use until
otherwise use before,
is to be sandwiched between two sub stevents, then
if the matrix event is to last for the whole of the period between the
two parts of the sub stevent, then for the first sub stevent, use from,
and for the second use to
unless the second is an event, then use until
otherwise use between ... and.
There are certain results that we have not shown in the selection tree, to
avoid clutter:
in with a non-definite time duration can also be used to indicate, not
a time period of that duration beginning at the TOD, but rather a
point in time which is located on the timeaxis at a point given by the
T O D plus this duration;
over, in expressions like over the weekend, is not shown. This use of
over is an alternative for at or in, depending on whether weekends etc.
are to be regarded as time points or time periods.
1.2. The selection tree for Dutch temporal function words

A similar rule holds for selecting Dutch temporal prepositions and conjunctions. Again the first and main distinction is whether the time of the
matrix stevent is being equated to that of the sub stevent or not. If it is,
then the nature of the sub stevent is considered. If the sub stevent is:
an event then:
if the sub event is a period, use terwijl
otherwise:
if the TOD is in the past and the matrix stevent is one off, then
use toen
otherwise use als (or wanneer);
a state, then:
if this state is a point in time, then:
if it is a clock time point, then use om,
if it is a event-time slice, then use op,
if it is at one end of an event-time box, then use aan,
299
unless the matrix stevent can take place at a time earlier than the
substate, then use tegen;
if the sub state has a duration then:
if the duration of the matrix stevent is to be equal to that of the
sub state, then use nothing, adding ... lang for emphasis,
unless the sub state is a period ending at the TOD, then use al,
or
unless the sub state indicates the purpose of the matrix stevent,
then use voor,
if the duration of the matrix stevent is to be less than that of the
sub state; then use binnen;
otherwise, use in unless the sub state is a time of day or a season
of the year, then use 's,
unless the sub state can also be given a non-temporal interpretation (usually locative), then:
if the sub state is also the cause of the matrix stevent, then
use bij,
if the sub state is incidental to the matrix stevent, then use
onder,
otherwise use tijdens.
If the time of the matrix stevent is not to be made approximately the
same as that of the sub stevent, then the order of the matrix and sub
stevents needs to be considered. If the matrix stevent is to fall in a period
that:
is
is
is
to follow the sub stevent, then:

if the other end of this period is to be the TOD, then use sinds
otherwise use na unless the sub stevent is an event, then use nadat;
to precede the sub stevent, then
if the other end of this period is to be the TOD, then use tot unless
the sub stevent is an event, then totdat may be used;
otherwise use voor, unless the sub stevent is an event, then voordat
may be used;
to be sandwiched between two sub stevents, then:
if the matrix stevent is to last for the whole of the period between
the two parts of the sub stevent, then for the first sub stevent, use
van, and for the second use tot
otherwise use tussen ... en.
There is one result that we have not shown in the selection tree, to avoid
clutter:
300
Martin Durrell -
David Bree
over, with a non-definite time duration, is used to indicate a point in

time which is located on the time axis at a point given by the TOD
plus this duration.
1.3. Periods versus temporal duration

One point in common to both the English and Dutch temporal function
words is that while most of them can only be used to indicate a point or
period on the time axis, a few can also be used to indicate a duration of
time. These few are some of the prepositions that are used to indicate
that the time of the matrix stevent falls within a time duration:
In English: in, within and for,
In Dutch: in, binnen and al.
The following rule for interpreting their function is:
if the preposition takes a definite noun phrase, e.g., for the summer,
then it indicates a time period already bound to the axis:
if the preposition takes an indefinite noun phrase, e. g . , f o r five minutes,
then it introduces a time duration. This duration may or may not be
attached to the time axis. If it is attached, then the duration can either:
begin at the TOD, or
end at the TOD, which is indicated by the matrix clause having a
perfect aspect.
We turn now to an analysis of the German temporal prepositions and
subordinating conjunctions. We considered in turn:
Simultaneous events;
Simultaneous points and periods;
Temporal ordering.
2. Simultaneous events
Table 1 summarises the translation equivalents found between German
and English conjunctions. First, some typical data for these are given.
301
2.1. Als = while, when, as

Als er in London war, nahm er Klavierstunden.
While/when (*as) he was in London, he took piano lessons.
Die Sonne schien an einem wolken-losen Himmel, als er seinen Heimatort
verlie.
The sun was shining in a cloudless sky as/when (*while) he left his home
village.
Als ich in Passau ankam, habe ich sie auf dem Bahnsteig gesehen.
When (?as, *while) I arrived in Passau, I saw her on the platform.
2.2. Wie = as, when

Wie ich an seinem Fenster vorbeigehe, hre ich ihn singen.
As (?when, *while) I passed his window I heard him singing.
Wie ich das Fenster ffne, schlgt mir heftiger Lrm entgegen.
When (*as, *while) I opened the window I was confronted by an intense
noise.
2.3. Whrend = while, when, as

Die Zollprobleme lste Boris, whrend wir in Urlaub waren.
Boris solved the problems with the customs while/when (*as) we were on
holiday.
Sie gingen weg, whrend wir uns unterhielten.
They left as/while we were talking.
2.4. Wenn = when

Ich bringe es, wenn ich morgen vorbeikomme.
I'll bring it when (*as, *while) I come tomorrow.
2.5. Sobald = as soon as

Das tat sie auch, sobald sie wieder nach Hause gekommen war.
She did that, too, as soon as she had got home.
302
Martin Durrell D a v i d Bree
2.6. The following should be noted in respect of these data

(i) In standard German wie is used with a present tense referring to the
past (so-called "historic present"). It is an (optional) alternative to als
with a past tense. The use of wie with a past tense, as an equivalent to
English as, is not uncommon but regarded as regional or non-standard,
cf. D U D E N (1976-81: 2877). In either case, the condition for its use is
the same as for English as, i. e., the sub stevent is an event which logically
follows, or is very close to, the previous TOD.
(ii) Although whrend may always be translated by while, it may be given
by as in certain contexts, principally where as refers to an activity.
Compare:
Ich las ein Buch, whrend er sich auszog.
I read a book as/while he was getting undressed.
Boris lste die Probleme, whrend wir weg waren.
Boris solved the problems while (*as) we were away.
This corresponds to the situation in Dutch, cf. Bree et al. (1990: 14-15).
(iii) In general, it seems to be the case that whrend and als are interchangeable in certain contexts, paralleling while and when in English, cf.
Bree et al. (1990: 12-13). However, it will need further investigation to
establish whether the conditions are exactly the same. In particular, native
speaker confirmation is needed on the acceptability of contexts such as:
?Als wir in Urlaub waren, hat Boris die Zollprobleme
gelst.
In general, we are dealing here in both languages with contextually

determined synonymy in special conditions.
(iv) The distribution of als and wenn precisely parallels that of Dutch
toen and als/wanneer, i.e., German als is limited to single events in the
past, whilst wenn is used for present, future (including future-in-the-past)
and repeated past (i.e., whenever she hit him ...), cf. Bree et al. (1990:
12-17). Like Dutch als, wenn is also the usual equivalent of English if in
conditional sentences, although in German the mood may disambiguate,
as the subjunctive is normal in sentences expressing unreal conditions,
cf. Durrell (1991 a: 317-321).
(v) English as may correspond to whrend or als, but NOT to wenn,
since as is not used in a temporal sense in present contexts. There is
considerable potential for contextual synonymy between whrend and als
303
Table 1.
German
English
whrend
while
wie/als
as
wenn
when
sobald
as soon as
as an equivalent for as, and this would merit further investigation. For
the use of wie, see 2.2.1.
We may draw the preliminary conclusion from these data that the
decision trees for German are almost identical to those for Dutch, not
least because als/wenn parallel toen/als. The only major difference is in
the (essentially optional) use of wie with the historic present. Thus:
to refer to point of time:
if a single occurrence in the past
unless present tense employed, optionally,
otherwise
to refer to a period of time
use
use
use
use
als
wie
wenn
whrend
3. Simultaneous points and periods

In practice, it was found useful in this initial survey to take all the
prepositions expressing simultaneity together, as the selection processes
in German turned out to be rather different to those described in Bree
et al. (1990) for English and Dutch. Table 2 shows the patterning of
"normal" (or "most usual") translation equivalences between the languages again with the proviso that the frequency and importance of
these correspondences has not been tested against translation data. The
304
Table 2.
German
English
gegen
around
um
zu
at
bei
an
innerhalb/
binnen
within
whrend
during
seit
rj auf
bis
for
by
until
nach
after
before
... ago
305
section below is ordered in terms of the German prepositions (or the use
of cases to indicate temporal relations), following the sequence of Bree
et al. (1990: 17-35) in terms of semantics.
3.1. Um = at
um vier Uhr
at four
o'clock
The use of urn parallels that of Dutch om precisely in that it sets the time
of the matrix stevent with clock times only. Thus the decision pathway
must involve the choices m = s, s is a state, a point in time and a clock
time, as with Dutch om.
3.2. Zu = at
zu Weihnachten, Ostern, Pfingsten
zu dieser Zeit
zur Zeit der letzten Wahlen
zu einer anderen Zeit
zu jeder Tageszeit
zu gleicher Zeit
zu dieser Stunde
zu jeder Stunde
at
at
at
at
at
at
at
at
Christmas, Easter, Whitsun

that time
the time of the last election
some other time
any time of the day
the same time
this hour
any time
Zu occurs:
a) in standard German with the major religious festivals (there is some
regional and register variation, cf. Durrell (1991b: 122)). This corresponds
very closely to the use of met in Dutch, which did not occur in the data
in Bree et al. (1990).
b) with the nouns Zeit 'time' and Stunde 'hour' in some contexts. The
use of zu or in with Zeit and Stunde is a notoriously difficult area, and
it may not always be consistent. This difficulty may be resolved, however,
if we assume that zu is used with these words if we are dealing with what
is perceived as a point in time rather than a period. In other words the
"time" or the "hour" is being presented as a unit, rather than as a
"container" within which another stevent occurs. This will become clearer
under in, cf. 3.8.
The decision tree for zu thus involves m = s, s is a state and a point
in time which is either a religious festival, Zeit or Stunde.
306
3.3. Gegen, um = round, about

It the time set for the matrix stevent is an approximation, two possibilities
are available in German, i. e.:
gegen = round, about, getting on for, towards
gegen halb drei
at about/towards half-past two
um round, etc.
Ich komme um Weihnachten (herum)
I'll come round about Christmas
There are certain limitations or restrictions on the use of gegen and

um, viz.:
i) gegen with clock times is notoriously ambiguous, and native speakers
disagree whether it means "at about" or "getting on for"/"approaching".
This is a matter of unpredictable idiolectal variation (and the source of
much argument, cf. the mutually contradictory information in D U D E N
(1976-81: 966) and Schrder (1990: 117)). With other than clock times
gegen clearly refers to a period immediately preceding, e.g.:
gegen Mittag
gegen Ende des Monats
towards midday
towards the end of the month
ii) The use of the intensifier herum in conjunction with um is essentially

optional. However, it is obligatory with approximate clock times, where
simple um indicates a point in time (i.e. = 'at'), cf. 3.1. above. Clearly,
this distribution of um ... {herum) is linked to the potential ambiguity of
gegen in an interesting way.
3.4. Bei = on, at during

bei
bei
bei
bei
Tagesanbruch
seiner Geburt
dieser Gelegenheit
der Probe
at daybreak
at his birth
on this occasion
during the rehearsal
The function of bei is clearly to indicate simultaneity with nouns which

do not themselves express temporality (i.e., it indicates 'at the time when
took/was taking/will take place'). In this respect it may refer to a
"point" or a "period", depending largely on the nature of the stevent
denoted by the noun governed by bei', bei must be used with events (or
307
activities/accomplishments) rather than states. In contrast to whrend, it

is only used as a preposition. If a period is involved, as in the last example
above, bei may be replaced by whrend. The distinction between bei and
whrend in such cases is convincingly explained by Schrder (1990: 217
8), who establishes that bei may be used only if m and s occur simultaneously and in the same place. If simultaneity alone is involved, then
only whrend is acceptable, cf.:
Whrend des Vortrags/Beim Vortrag ist es unhflich zu sprechen.
It is not polite to speak during the lecture.
Bernd kam noch whrend (*bei) meiner Abwesenheit an.
Bernd arrived during my absence.
Furthermore, as Schrder (loc. cit.) also implies, it is generally the case
that, with bei, the stevent is envisaged as a point, whilst whrend generally
implies duration. However, this would mean that, e. g., bei der Probe and
whrend der Probe would not be wholly synonymous. Schrder offers no
evidence which would justify this further assumption, which would require
additional corroboration. Thus, the decision tree for bei will include
m = s, and s is a simple non-temporal event.
3.5. An = on, in, during, at, #
An is the most wide-ranging of the temporal prepositions in German:

am Dienstag
am Morgen
an diesem Tag
an Wochentagen
am 5. Mai
am Tag
am Anfang
am Ende
am nchsten Dienstag
on Tuesday(s)
in the morning
that day
on weekdays
on the 5th of May
during the day
in the beginning
at the end
next Tuesday
In the terms used in Bree et al. (1990: 20), an is characteristically used

with time "slices", i.e., beginnings and ends, and also all days and parts
of days (except Nacht, where, like English, there is an idiomatic difference
in the preposition used, cf. in der Nacht 'at night'). The crucial determinant
for the selection of a preposition in the various languages under consid-
308
Martin Durrell -
David Bree
eration would appear to be what is perceived as a 'slice' in the specific

language. It could be possible to see an as introducing one of a series,
but its use does not correspond absolutely with that of English on, whose
major function this is. There would seem, though, to be no need in
German to adduce "two dimensional event-chunks", as Bree et al. (1990:
20) do for Dutch aan, to identify the use of an in German.
Notes on the use of am
(i) As in Dutch and English, the use of a preposition is not always
obligatory in contexts such as those given. This is to a certain extent a
matter of register (the omission being more frequent in the colloquial
language), or regional provenance.
(ii) Anfang and Ende are used without a preposition if followed by a
further time specifier, e. g.: Anfang/Ende Mai, AnfangIEnde 1992, Anfang/
Ende nchster Woche. (Mitte also follows the same pattern, e.g., Mitte
Januar). In such contexts we may be dealing with a duration rather than
a point, cf. the corresponding Dutch pattern of begin maart, eind januari,
half augustus.
Thus, the path for the selection of an involve m = s, s is a state, s is
a point and s is an event-time slice.
3.6. Genitive case = in

eines Tages
morgens (older: des Morgens)
one day
in the mornings
There are two rather distinct uses of the genitive, i. e., (i) to indicate an
indefinite, restricted point in time {eines Tages), and (ii) to indicate
repetition (morgens). Both are rather idiomatic in modern German and
restricted to certain lexical items and forms, so that it could be felt more
appropriate to deal with them as isolated adverbials rather than as
reflecting a productive use of the genitive.
Notes on the use of the genitive case in time adverbials:
(i) am Abend and abends, etc. may be alternatives to one another, such
that Abends gehe ich mit dem Hund spazieren and Am Abend gehe ich mit
dem Hund spazieren are synonymous alternatives, meaning either 'In the
evenings I go for a walk with the dog' or 'This evening I am going for a
walk with the dog'. It is not clear whether there are any contextual or
pragmatic restrictions on this synonymy, but it is paralleled in Dutch,
309
where 's avonds may similarly refer to habitual or repeated action or a

single point, cf. 's Avonds ga ik een wandeling doen met de hond.
(ii) Unlike Dutch, seasons cannot be used in genitive constructions in
German, cf. in in section 3.8. It should be noted, though, in respect of
Dutch, that this construction can only be used with the basic 'Germanic'
seasons, i. e., 's zomers and 's winters (or: in de zomer, in de winter) but
not * 's herfsts, *'s lentes.
3.7. Bis = by, until/till

Bis heute abend habe ich es fertig.
I'll have it finished by tonight.
Ich arbeite hier bis heute abend.
I'm working here till tonight.
From the point of view of English, bis is notoriously ambiguous, as the
examples show. It can indicate:
a) that the matrix stevent must occur between the T O D and the sub
state;
b) that the sub state is a termination point for the matrix stevent which
is commencing at TOD.
The matrix stevent in (a) is punctual, but in (b) it is durative, and what
is happening is that German is not distinguishing this lexically. Seit has
a similar range, cf. 3.13. It would be dubious to try to establish distinct
lexemes here, i.e., bis1 and bis2 to correspond to English by (or Dutch
tegen) on the one hand and English until (or Dutch tot) on the other.
D U D E N (1976-81: 395) simply defines temporal bis as "gibt die Beendigung eines Zeitabschnitts an" (i. e., it indicates the end of a period of
time). What bis does is to indicate a period between the T O D and the
sub state, and the matrix stevent occurs within that period, whether it
lasts over the whole period ( = 'until') or falls at an unspecified point
within it ( = 'by'). It would be difficult to accommodate this within the
decision tree given in Bree et al. (1990: 43-44), as the first decision is
+ (m = s), and bis would need to appear on both sides of this, i.e., in
the places of both English by and until. A solution could be to reanalyse
English and Dutch and move by (and Dutch tegen) from the m = s to
the ~ ( m = s) pathway, as they are concerned with temporal ordering
rather than indicating simultaneity. In this way the range of meaning of
bis could be coped with in the same framework as operates in Dutch and
English.
310
3.8. In = at, in, during, #

in diesem Augenblick
im folgenden Monat
im kommenden Jahr
im Sommer
im Mai
in der Zeit nach dem Krieg
in der Nacht
in den sechziger Jahren
im Krieg
at that moment
the following month
next year
in the summer, during the summer
in May
in the time after the war
at night, during the night
in the sixties
during the war
Clearly, in is the "container" word, although it is not easy to visualise

the distinction between this and "borders" or "slices" (i. e., an). The use
of one or other prepositions with particular words or sematically related
groups of words (e.g., "days of the week", etc.) seems partly arbitrary
or a matter of language-specific perception, if not purely collocational.
Compare, for example: at that moment, op dat ogenblik, in dem Augenblick.
The differences here seem to depend on how the speakers visualise
"moments", whether as "points" or "having duration". We note that
German, which uses its container word in in this context, is quite happy
to use lang with Augenblick, e.g., einen Augenblick lang, cf. Bree et al.
(1990: 32), which points out that lang is not used with ogenblik in Dutch.
It would thus appear that, in German, "moments" are perceived as having
duration, whereas in Dutch they are interpreted as "points".
Again, like an, in may be omitted in some contexts, e. g., (im) folgenden
Monat. The constraints on this are not wholly clear.
Where in is used with Zeit and Stunde, it would appear to be the case
that they are being visualised as "periods" rather than "points", e. g.:
In der ganzen Zeit haben wir sie nicht
In all that time we didn't see her.
gesehen.
In unserer Zeit tut man das nicht mehr.

In our times that is no longer done.
in ruhigen Stunden
in peaceful hours
Compare the use of zu with these words given in 3.3.
German in has the same range of ambiguity as English in, i. e., there
is nothing corresponding to Dutch over, e.g.:
311
Ich komme in einer Woche.

I'm coming in a week.
in vier Jahren
in four years (time)
in fnf Minuten
in five minutes
This polysemic range is not dissimilar to that of German bis and can be
coped with without postulating distinct lexical items. Essentially, German
in is selected by m = s, s is a state having duration and duration of
m = duration of s.
3.9. Innerhalb = within

innerhalb von fnf Minuten
within five minutes
This is a hyponym of in, used to resolve the ambiguity identified in 3.8.

if necessary and indicating clearly that duration of m < duration of s.
In formal registers binnen is a possible alternative to innerhalb without
any significant semantic distinction, being in origin a regional alternative.
3.10. Whrend = during

The use of whrend as a preposition ( = 'during') is clearly linked with
its use as a conjunction ( = 'while', 'whilst', cf. 1.2.2.), and these uses
may most appropriately be brought together in the decision tree.
whrend der Auffhrung
during the performance
Whrend des Krieges lebten sie im Ausland.
During the war they lived abroad.
The distinction between whrend and in is that whilst both have duration of m < duration of s, with whrend s may be non-temporal. This
parallels English and Dutch. It is, however, notoriously the case that
whrend is less frequent than during, and this needs further investigation.
In particular, it appears that during may be more readily used with
temporal words than may whrend, cf.:
312
am Tag
in der Nacht
during the day

during the night
Schrder (1981: 110) asserts that whrend is restricted to use with nouns
which do not express temporal concepts, cf. * whrend des Morgens,
*whrend der Nacht and is in this respect similar to bei, cf. 3.4. However,
this is not true without further reservation, cf. Schanen Confais (1986:
458):
Whrend der letzten Nacht ist der
Kleine zweimal aufgewacht.
During the previous night the

little boy woke up twice.
Or, cf. D U D E N (1976-81: 2833):

Whrend dreier Jahre verbrachten sie den Urlaub in Sylt.
Three years running they spent their holiday in Sylt.
A possible explanation is that whrend expresses a period, not a duration,
within which a stevent occurs, and it is barred from occurrence in contexts
which, as with unqualified time words, it might be open to interpretation
as a duration. But the restrictions on the use of whrend require further
investigation.
3.11. Accusative case = for, #

Die Schule blieb eine Woche geschlossen.
The school stayed shut for a week.
Den ganzen Tag lag er im Bett.
He stayed in bed the whole day.
Ich war einen Monat (lang) in Kassel.
I was in Kassel for a month.
The accusative of time, without a preposition, is employed where the
duration of m = duration of s and there is no explicit link to the time
axis, in particular not to the TOD. We would maintain that this is the
case even with a present tense verb, as here a time adverbial is necessary
to make the link to the time axis clear. Thus, a sentence like:
Ich bleibe den ganzen Abend zu Hause.
I'm staying at home the whole evening.
313
could be understood as referring to "today" or "tomorrow" (or any other

evening in the near future) unless disambiguated by pragmatic factors or
by an explicit adverbial such as heute 'today' or morgen 'tomorrow', etc.
Notes on the use of the "accusative of time"
(i) As in Dutch, lang may be added for emphasis, and there is no real
knowledge of the motivation for the use of lang, nor of any possible
constraints on this use. It may, unlike Dutch, even occur with Augenblick,
cf. 3.8.
(ii) The accusative is a frequent alternative to the prepositions in, an or
aufjfr under conditions which need further investigation.
3.12. Fr/auf = for

Ich fahre (am Montag) fr) auf ein paar Tage in die Schweiz.
(On Monday,) I'm going to Switzerland for a few days.
Frlauf are used if the duration of m = duration of s and the period
starts either at a given future point or, by default, at TOD. The accusative
is an occasional alternative, mainly in colloquial speech. Schmitz (1964:
17) asserts that the temporal use o f f u r presupposes some idea of purpose,
as Bree et al. (1990: 33) also suggest for Dutch voor in similar contexts.
However, Schrder (1981: 113) shows with a number of examples that
this is not the case, and no consistent distinction can be identified between
a phrase in the accusative and one with fr on this basis. Any notion of
purpose in sentences containing fr is derived from other elements in the
sentence rather than from the contrast between fr and an accusative.
Thus, we should regard sentences like the following pair as registerdependent alternatives:
Ich gehe jetzt eine halbe Stunde ins Cafe.
Ich gehe jetzt fr eine halbe Stunde ins Cafe.
I'm going down the pub for half an hour.
In general, auf may most plausibly be seen as a rather more formal
alternative to fr, despite the assertion to the contrary by Schmitz (1964:
58). However, the exact distribution of these prepositions is not absolutely
clear and requires further investigation on the basis of a sufficiently large
and reliable body of data.
314
3.13. Seit = for, since

Ich arbeite seit vier Wochen hier.
I've been working here for four weeks.
Ich arbeite seit Ostern hier.
I've been working here since Easter.
No more than with bis does it seem appropriate to think in terms of
distinct "meanings" for seit, which indicates:
(a) a substate of given duration ( = m) ending at the TOD;
(b) that m continues over a period from s to the TOD.
Common to both uses is the notion of "up-to-now", i. e., the marking
of a range of time ending at TOD, and seit may be followed by an
indication of the duration or by the starting point of s. This range of
meaning is not easy to accommodate within the decision trees given in
Bree et al. (1990). A possible solution, as with bis, could be to link up
English since and for (and Dutch sinds and al) in terms of m = s.
However, one must then consider whether since actually does involve
temporal ordering, or whether it expresses the simultaneity of m and s.
At all events, the similarity of this case to that of bis suggests that there
may be an interesting generalisation to be drawn from the behaviour of
bis and seit and their English equivalents.
3.14. Vor = before, ago

Ich komme vor Weihnachten.
I'm coming before Christmas.
Ich war vor zwei Jahren hier.
I was here two years ago.
Vor, too, is ambiguous from the point of view of English, and the sense
'ago' does not occur in the data of Bree et al. (1990), because it is not
expressed by a preposition in Dutch or English (cf. Dutch twee jaar
geleden). In examples like the first we are dealing with temporal ordering,
in that m falls in a period preceding s. It corresponds to English before
and Dutch voor although it may be noted that the latter can be
multiply ambiguous. Compare:
Ik ga er voor twee weken naar toe.
I'm going there for three weeks.
315
Kommt hij inderdaad voor half zeven?

Is he really coming before half past six?
In examples like the second, m is related to the TOD; vor indicates a
point on the time axis prior to the TOD. It is the converse of in or of
Dutch over, which is omitted from the selection tree in Bree et al. (1990).
Both uses of vor involve the specification of a period prior to a point of
reference in the first case given after vor, in the second the TOD. It
would seem desirable to conflate them in the decision tree.
4. Temporal ordering
A characteristic of English, which is only partly shared with German
(and quite distinct from Dutch), is that many of the lexemes which
express temporal ordering (e. g., after, before, since) may be used either
as conjunctions or prepositions. It seems clear, if theoretically rather
interesting, that this distinction is a relatively low-level one, made at a
relatively late point on the decision tree. However, as German tends to
make the distinction (although the individual lexemes are often derivationally related), it is convenient to present them separately in the first
instance. In general, temporal ordering involves fairly straightforward
equivalents between German and English.
4.1. Prepositions
nach = after, later
nach vierzehn Tagen
after a fortnight/a fortnight later
(indicating the beginning of a range in which m occurs relative to TOD)
zwischen ... und between ... and
zwischen sieben und acht
between seven and eight
(m sandwiched between two sub stevents)
von ... bis = from ... to
von acht bis neun
from eight to nine
(m lasts over whole period between two sub stevents)
316
von ... (an)/ab = from ... (on)

von 1975 (an), ab 1975
(giving starting point for s)
from 1975 (on)
4.2. Conjunctions
nachdem = after
Die Probleme huften sich, nachdem wir die Grenze passiert hatten.
The problems mounted up after we had crossed the frontier.
bevor = before
Ich will das Fenster reparieren, bevor ich in Urlaub gehe.
I want to fix the window before I go on holiday.
NB: ehe is synonymous with bevor, it is restricted to literary and formal
registers.
bis = until, by the time
Ich warte hier, bis du zurckkommst.
I'll wait here until you get back.
Bis du zurckkommst, habe ich das Fenster repariert.
I'll have fixed the window by the time you get back.
seit(dem) = since
Er wohnt in einem Hotel, seit (dem) er sein Haus verkauft hat.
He's been living in a hotel since he sold his house.
The shorter form seit, homonymous with the preposition, is now more
frequent than the traditional conjunction seitdem, despite the assertion
to the contrary in Buscha (1989: 98). It does not have the semantic range
of the corresponding preposition, presumably because that range is not
compatible with an event.
5. Designing the selection tree for German temporals

While it would have been possible to use the selection trees for English
and Dutch temporals, as found in Bree et al. (1990), for the German
temporals, we have seen in the above analysis that this would have led
to some difficulties:
317
whrend would have appeared in two places, once instead of while and
once instead of during. These are separated too early in the English
selection tree.
bis would have occurred in two places, once instead of by and once
instead of until. These are separated at the first decision point in the
English tree!
We have decided to make changes to avoid these double occurrences.
5.1. Differences with the English selection tree

In the first place the distinction between event and state when the sub
stevent sets the time for the matrix stevent, found in the English and
Dutch trees (top half), will be dropped. This distinction turns out not to
be felictious. In the English and Dutch trees it corresponded to the use
of a subordinating conjunction (event) versus a preposition (state). However, subordinating conjunctions can be used to introduce states as well
as events, e. g.:
When the rod is white hot, it is malleable.
Further, the distinction between prepositions and subordinating conjunctions hardly exists in the other (bottom) half of the selection trees,
i. e., when the sub stevent is used to set one end of a time range for the
matrix stevent. So in the m = s branch this distinction will be placed
lower in the German selection tree. As whrend can be used both as a
preposition and a subordinating conjunction, no distinction will have to
be made when the sub stevent is a period. When the sub stevent is a
point, then the distinction has to be made immediately.
Another anomaly in the English and Dutch selection trees was the
placing of by and tegen in the top half of the tree, in which the matrix
stevent time is set to that of the sub stevent time. Rather, for both by
and tegen the matrix stevent should occur sometime before the sub stevent
time. In German bis is used for this. Bis is also used for the main stevent
occurring for the whole of the period from the T O D to the sub stevent
time. In English this is the function of until, which is in the other half of
the tree from by, and similarly for Dutch tot and tegenl We will place
both situations in the same half ( ~ M = S) of the German tree.
A related issue is the place of English for. Currently it is shown as
signifying that m = s, and so is in the "top" half of the tree, and that
the duration of m = duration of s. However, its equivalents in German
318
are the simple accusative unless the period of m is related to the TOD.
If m is related to the TOD then if m ends at the TOD, the German uses
seit (cf. Dutch al) and if it begins at the TOD then it uses fr or auf.
Now seit is also used for English since which occurs in the "bottom"
( ~ m = s) half of the tree! So once again there is an indication to move
for from the top (m = s) half to the bottom ( ~ m = s) half of the tree.
However, when German uses the accusative for a pure duration, as in:
Ich war einen Monat in Kassel.
I was in Kassel (for) a month.
then there would seem no reason to place this in the bottom ( ~ m = s)
half of the tree.
The situation when the sub stevent is a pure time duration does not
lend itself to the same decision criteria as are suitable for the case when
the sub is really a point, period or event. This is so, despite the fact that
the same temporal prepositions are used! This is an unhappy state of
affairs, which needs to be resolved. We give rules for the selection of a
preposition with pure duratives in 5.3.
The simple distinction found in the English tree for m = s when s is a
state, viz a choice between the sub stevent being a point, from a series
or having duration, is not sufficient for German, just as it was not for
Dutch. However the solution used for Dutch is also unsatisfactory since
some prepositions, e. g., al, binnen, would be better placed in the other
half of the tree. We propose that the first decision in the top (m = s)
half of the tree should do most of the work. It as at this level that not
only points, days and periods should be distinguished but also duration
and simple events, i.e., events expressible without using a sub clause.
This first decision point will be on the nature of the sub stevent: is it a
point, a (part of a) day, a pure duration, a simple event or a period.
In the lower ( ~ m = s) half of the tree, the only change is the introduction of frIauf and bis (as equivalents for English for and by ). The
two uses of seit (in English since and ... ago are naturally captured. It is
the English tree which needs adjusting.
We are now in a position to propose the selection tree for German
temporals. To have an overview from which to work, we present in
summary form the information from the previous sections, ordered by
the role that the sub stevent is playing in setting the time of the matrix
stevent (see Table 3).
319
Table 3. Quick index to English, German and Dutch temporals, by type of sub stevent
Description
S is a duration
Cycle time of = "every" S
Duration of = S:
is a state/activity:
& < TOD
& TOD >
is an accomplishment
related to TOD:
< TOD + S
= TOD + S
= TOD S
= now S
S alone locates on time axes
S is a point:
Clock time
"the beginning/end"
"the beginning/end of ..."
S is an event:
Simple point
Complex:
once in past
+ historic present
otherwise
S is a whole/part of a day:
Simple part, e.g., morning
Qualified part, e.g., fine morning
Definite day, e.g., Sunday
Indefinite day e.g., one Sunday
S is a period:
Simple
Complex
English
German
Dutch
# + acc
#
#
if1for
for
for
in
# +acc
seit
fr/auf
in
al
#
within
in
... before
... ago
innerhalbIbinnen
in
... zuvor
vor
binnen
over
... geleden
... geleden
at
at
at
um
an
# +acc
om
aan
# +a.cc/aan
at
bei
when/as
als
wie
wenn
bij
when/as
toen
als/wanneer
in
on
#/on
#
# +gen/an
# + gen/an
in
op
*!op
op
in/during
while
in/whrend
whrend
in/tijdens
terwijl
falls in a period, one of more ends of which are indicated by S

is located on semi time axis:
S <
after
nach (dem)
< S
(be)vor
before
is located in a time range:
5 < < TOD
seit (dem)
TOD < < S:
6 duration of = S - TOD
until/till
bis
& occurs once
by
bis
SI < < S2:
& duration of = S2 - SI
from ... to/until
von ... bis
& SI < < S2
between ... and
zwischen ... und
na( dat)
voor( dat)
sinds
tot (dat)
tegen
van(af) ... tot
tussen ...en
320
5.2. The German selection tree

The first distinction to be made is the well known one between the sub
stevent setting the time of the matrix stevent, or setting one or more ends
of the matrix stevent. In the latter case the other end of the matrix stevent
is either open, set by the T O D or by another sub stevent. For the case
when the sub stevent sets the time of the matrix stevent we have first to
decide on the type of sub stevent:
If the sub stevent is a point, which is
a clock time, then use urn;
a religious feast, or a phrase using Zeit or Stunde referring to a point
of time, then use zu\
represented by a clause:
if the sub clause indicates a single event in the past, then use als,
unless the historic present tense is being used in the matrix clause,
in which case use (optionally) wie;
otherwise, use wenn.
If the sub stevent is a (part of a) day of the week, or "the beginning/
end" {AnfangjEnde) with no further specification of what is beginning
or ending, then use an.
If the sub stevent is a period:
if the sub is NOT a phrase with a temporal term, then use whrend,
unless the sub stevent is an action, then use bei.
otherwise use in, unless the duration of the matrix stevent is to be
marked unambiguously as less than the duration of the sub stevent,
then use the accusative, with optional lang.
For the case when the sub stevent marks one or more ends of the time
in which the matrix stevent occurs, we have first to decide on the order
of the matrix and sub stevents:
If the matrix stevent occurs later than the sub stevent, then:
if the T O D is an upper bound on the time range for the matrix
stevent, then use seif,
otherwise use nach, or, if the sub stevent is a clause, nachdem.
If the matrix stevent occurs earlier than the sub stevent, then:
if the lower bound on the time range for the matrix stevent is the
T O D then use bis, unless the sub stevent is a pure time duration
321
indicating the time prior to the TOD in which the matrix stevent
occurs, then use fr or auf (cf. 5.3. below);
otherwise use vor, or, if the sub stevent is a clause, bevor.
If the matrix stevent is to occur in a range given by two sub stevents
s1 and s 2, then:
if it is to endure for the whole of the range from s i to s 2 then use
von s 1 bis s 2;
otherwise use zwischen s 1 und s 2.
These rules are represented in Figure 3.
5.3. A note on the use of temporals with pure time duration

When the sub clause or phrase gives a pure time duration not connected
to the time axis, e. g., drei Monate 'three months', then the rules in the
bottom half of the selection tree are not directly applicable. The data to
be represented are shown in Table 3. There are three main cases:
The sub stevent is the cycle time of the main stevent, e. g., alle paar
Jahre every few years, in which case there is no preposition and the
accusative case is used.
The duration of the matrix stevent is equal to that of the sub stevent,
then:
if the matrix is an activity or state, then:
if it is to take place for S time before the TOD, then use seit;
if it is to take place for S time after the TOD, then use fr or auf,
otherwise, use the accusative with no preposition
if the matrix stevent is an accomplishment, then use in.
matrix stevent occurs at a time related to the T O D by s:
within a time S after the TOD, use innerhalb or binnen
at a time S after the TOD, use in;
later than a time S after the TOD, use nach
at a time S before the TOD, use vor.
These rules are depicted in Figure 4.
What is surprising is that this selection tree "looks" so different from
that used for selecting a temporal with a non-durative, as shown in Figure
3, yet the prepositions are largely the same! There must exist a better
perspective that would make this overlap obvious.
322
uses
historic
present
als
Sub stevent
is one off
in the past
-
a clause>
a clock time
a religious feast, Zeit, Stunde
a (part of a) day,
the beginning/end
wahrend
Sub Steven
is an action
Sub includes
temporal
term
bei
in
=S
#+acc (lang)
M= s
nach(dem)
< TOD
+
seit(dem)
S<M/
7wisrhrn ... nnrt
?S
S1<M
M<S2*
M=
[S1.S2]
+
vnn
, bis
M<S
TOD <
(be)vor
M, S Time of Matrix/Sub Stevern
, S Duration of Matrix/Sub Stevern
TOD Time Of Discourse
+ Condition fulfilled
- Condition NOT fulfilled
Default
Figure 3. Selection tree for German non-durative use of temporals
Duration o f Sub
TOD
323
Time O f Discourse
Default
Figure 4. Selection tree for German durative use of temporals
6. Conclusion
We have seen that there is much similarity between the German use of
temporals on the one hand and the English and Dutch on the other.
However, there are some differences which are sufficiently large to require
a quite different selection tree for the German prepositions. In particular
in German:
there is one word, whrend, that is both a preposition and a subordinating conjunction, whereas English and Dutch have separate words
(during I while and tijdensjterwijl). This has forced the point for selecting
between a temporal preposition and conjunction further down the
German tree;
there is one word, bis, which indicates the period from the TOD to S
and does not distinguish whether the matrix stevent occurs for the
whole of this period or simply within it (cf until/by and totjtegen). To
be consistent with the German tree, it would be necessary to alter the
324
English and Dutch trees to make by and tegen be in the bottom

( ~ m = s) halves, which would improve them;
there is one word, seit, for the time period and the time duration before
the TOD (cf since /for and sindsj ... ). Again, in order to make the
English and Dutch trees to be consistent with the German tree would
require these words to be moved from the top to the bottom halves of
the trees, which is possibly desirable;
there are three words for indicating time durations, depending on
whether or not the duration is before the T O D {seit), after the TOD
(fr/auf) or not related to the TOD (simple accusative), whereas there
is only one word in English (for) and none in Dutch.
It has become clear that the selection trees proposed for English and
Dutch could not select properly when the sub stevent was a pure duration.
The greater number of words that accept pure duratives in German and
the use of the same prepositions as for the non-duratives, has brought
home that the selection criteria in the English and Dutch trees were not
adequate for dealing with duratives. Two selection trees have been proposed for German, one for non-durative use and one for duratives. That
almost all the temporals which occur with duratives are also used with
non-duratives, means that we loose parsimony and break the design rule
that each proposition must occur only once in the selection tree. However,
pending a solution for incorporating both uses into one tree, we have
been forced to present two trees.
On the positive side, the criteria used for the English and Dutch
selection trees have been reused for the German tree. They have made at
least one point of the German use of temporals clear, namely that Zeit
and Stunde expressions could be used either with zu or with in. The
resolution is that this depends on whether the time or hour is regarded
as a point (zu) or as a period (in). Moreover, the analysis of the German
use of temporals has thrown light on the English and Dutch use and will
lead to a revision of their selection trees.
The outstanding questions for the German selection tree, aside from
how to combine the durative and non-durative use of the same temporals
are:
to have a better distinction between in and whrend as a preposition;
to find a distinction between fr and auf, if one exists.
325
Notes
1. From the Departments of German and Computer Science of the University of Manchester, U. K.
References
Bree, David Sidney Ruud Smit Jacques von Werkhoven
1990
"Translating temporal prepositions between Dutch and English", Journal of
Semantics 7: 1-51.
Buscha, Joachim
1989
Lexikon deutscher Konjunktionen. Leipzig: Verlag Enzyklopdie.
DUDEN
1976-1981 Das groe Wrterbuch der deutschen Sprache. 6 vols. Mannheim: Dudenverlag.
Durreil, Martin
1991 a
Hammer's German Grammar and Usage. (Revised edition.) London: Edward
Arnold.
1991 b
Using German. Cambridge: Cambridge University Press.
Oxford-Duden
1990
The Oxford-Duden German Dictionary. Oxford: Clarendon Press.
Schanen, Francois Jean-Paul Confais
1986
Grammaire de l'allemand. Formes et fonctions. Paris: Nathan.
Schmitz, Werner
1964
Der Gebrauch der deutschen Prpositionen. Mnchen: Max Hueber Verlag.
Schrder, Jochen
1981
"Temporaladverbien und Prpositionen", Deutsch als Fremdsprache 18: 106115.
1990
Lexikon deutscher Prpositionen. (2nd edition.) Verlag Enzyklopdie. Leipzig.
Terrell, Peter Veronika Schnorr Wendy Morris Roland Breitsprecher (eds.)
1990
Collins German Dictionary. London and Glasgow: Collins.
On universality and variability in the semantics of

spatial adpositions
Bruce W. Hawkins
1. The problem
Recent work in cognitive linguistics has revealed a great deal about the
nature of semantic structures in particular languages. Because the most
influential works in this emerging field of linguistic analysis reject the
fundamental assumptions of the Chomskyan research program (e.g.,
Langacker 1987, Lakoff 1987), the issue of universality of these semantic
structures has not been confronted with any vigor by cognitive linguists.
I find this problematic an instance of throwing the baby out with the
bath water. Whatever the nature of the structures under investigation,
those structures must be acquirable. And as soon as we confront language
from the acquisition perspective, the question of universality arises
whether we like it or not.
The underlying purpose of this paper is to re-introduce the issue of
universality into the research program of cognitive linguistics. I will do
so through an examination of semantic structures in the domain of spatial
relations. I will confront most directly the question of what semantic
structures characteristic of English spatial prepositions may be considered
universal. Derivatively, I will also consider the question of what factors
lead to observed variability across languages of these supposedly universal
semantic structures?
These questions will be confronted within the framework of Langacker's Cognitive Grammar (CG). CG assumes an encyclopedic perspective
on meaning, asserting that the meaning of any particular linguistic unit
is inextricably tied to a complex web of linguistic and extralinguistic
contexts. The substantive role of context in semantic structure is captured
in CG by means of the profile/base distinction. The base in semantic
structure constitutes a composite of specific and generalized contexts
which make up the cognitive background relative to which a particular
linguistic unit is semantically characterized. The entity highlighted in a
particular semantic structure is the profile.
328
Bruce W. Hawkins
Langacker (1987: 149) touches briefly on the issue of universality in

his discussion of the base in semantic structure. He claims that every base
involves specifications in one or more cognitive domains and that some
of these domains may be conceptual primitives, hence universal. He refers
to these as basic domains. He suggests that the search for universals
among the inventory of profiled structures is outside the scope of CG.
Thus, Langacker restricts our research on semantic profiles to the question of what structures are profiled. This paper takes issue with this selfimposed limitation on the CG research program. I propose that it is both
necessary and possible to confront productively the universalist question
of what structures are profilable.
Langacker actually began the process of investigating this latter question years ago by proposing a fundamental subcategorization within the
set of profiled semantic structures. He identifies one category of profiles
characteristic of nouns and another characteristic of verbs, adjectives and
prepositions. The former are referred to as THINGS. Any semantic
structure identified as a T H I N G presents a profile characterizable as a
region in some domain.
Herein, we are concerned most directly with a subset of the latter
category, referred to in CG as RELATIONS. The central feature of any
relational profile is a cognitive link (cf. Johnson 1987: 117-118) or
interconnection. The entities linked may be T H I N G S or RELATIONS.
The typical RELATION involves a linkage of two entities, but three or
more entities may be involved. In any case, there will always be one
particular entity in a relational profile that is most prominent. This entity,
which stands out as a figure in the relational profile, is referred to in CG
as the TRAJECTOR (TR). Any other entity in the relational profile
constitutes a L A N D M A R K (LM) which provides a salient point of
reference for situating the TR within a particular cognitive domain.
In this paper, I will examine the inventory of spatial relations profiled
by English prepositions and Japanese postpositions. My ultimate objective is to provide not just an inventory of profiled spatial relations, but
more ambitiously, a theory of why the inventory includes the relations it
does. As a result, we confront the question of universality head on. That
is, the central question becomes not what are the profiled relations but
rather what relations are profilable?
On universality and variability of spatial adpositions
329
2. The semantics of English spatial prepositions

Adequate description of the set of semantic structures profiled by English
spatial prepositions necessitates three fundamental parameters of semantic structure: basic relations, configurations instantiable by the prepositional trajector (hereafter TR configurations), and configurations instantiable by the prepositional landmark (hereafter LM configurations). I
shall begin substantive discussion of this framework with the basic relations.
2.2.1. Basic relations
There are only two basic relations which are profiled by spatial prepositions in English, COINCIDENCE and SEPARATION. In other words,
the specifically relational content of any spatial preposition in English
can be identified as either COINCIDENCE or SEPARATION. Each of
the prepositions apparent in (1) profile the COINCIDENCE relation,
while the contrasting prepositions in (2) exhibit the SEPARATION relation.
(1)
a. Erika is at home.
b. Erika is on the ugly blue carpet.
c. Erika is in her bedroom.
(2)
a. Erika is away from home.

b. Erika is off the ugly blue carpet.
c. Erika is out of her bedroom.
The essential defining characteristic of COINCIDENCE is identified

explicitly in the principal definition of the verb "coincide" given by
Webster's Third New International Dictionary: "to occupy the same place
in space". The crucial word here is "same"; all other terms in the definition
simply indicate that the relation involves location in space. The word
"same" identifies the quality of this spatial relation spatial cooccurrence, two entities perceived as occupying one physical space. In contrast,
the defining characteristic of SEPARATION is difference. In any case of
SEPARATION, there are two (or more, perhaps) entities perceived as
occupying two different physical spaces. A natural consequence of this
is that it is possible, indeed a logical necessity, to perceive a spatial gap
between the entities in a SEPARATION relation. No such spatial gap
can be perceived in a COINCIDENCE relation.
330
Bruce W. Hawkins
There are few locative prepositions in English, outside of those listed

in (1), which profile a COINCIDENCE relation. Indeed, the only other
reasonably uncontroversial members of this set are the prepositions in
(3).
(3)
a. Ken is beneath the covers.

b. Ty is underneath Ken (screaming at the top of his lungs).
c. Erika is on top of this small mass of humanity.
In the case of the prepositions in (3), as opposed to those in (1), the

COINCIDENCE relation must be profiled against a cognitive background which makes the vertical dimension identifiable. There appear to
be no prepositions in English which profile COINCIDENCE specifically
in either of the horizontal dimensions. As shown in (4), however, there
are prepositions profiling SEPARATION in all three dimensions in space.
(4)
a.
b.
c.
d.
e.
f.
g.
h.
The plane is above the clouds.

The plant is over the kitchen sink.
The plane has dropped below the clouds.
You'll need to put a bucket under that hole in the ceiling.
The blue bus parked in front of my car.
The yellow bus parked behind/in back of my car.
My chair is to the left of the bookcase.
The stereo is to the right of the bookcase.
SEPARATION relations may also be distinguished on the basis of

distance of the SEPARATION. In most cases in English, this is achieved
with a prepositional qualifier rather than by the preposition itself. This
is the case in (5).
(5)
a. The plane is far above the clouds.

b. The gate is 10 meters to the left of the house.
c. The supermarket is only one mile down the road.
However, there is at least one preposition in English, near, which

profiles a PROXIMAL SEPARATION relation.
(6)
The Normal Post Office is near the ISU campus.
When we turn our attention to directional prepositions, we find that

most profile a COINCIDENCE relation. For each preposition in (7),
there is COINCIDENCE between the LM configuration and some restricted portion of a linear TR configuration.
(7)
a. John jumped into his car.

b. John backed out o/his garage.
331
c. John drove through the intersection without stopping for the

red light.
d. John drove from Chicago to Los Angeles in record time.
In the case of into and to, COINCIDENCE involves only the terminal
location, or TERMINUS, of the linear TR configuration. By contrast,
the COINCIDENCE profiled by out of and from involves only the initial
location, or ORIGIN of the linear TR configuration. In the example of
through in (7 c), the COINCIDENCE relation involves some non-initial,
non-terminal location or set of locations. In (8), on the other hand, we
see that the same preposition can profile a COINCIDENCE relation
which involves the entire linear TR configuration.
(8)
He waved his sword defiantly through the air.
Similarly, the prepositions in (9) also exhibit COINCIDENCE which is

exhaustive of a significantly expansive TR configuration. In these cases,
however, the expansiveness is clearly not linear.
(9)
a. There is smoke throughout the library.

b. There are papers all over my desk.
c. There is dirt all across the road.
2.2.2. TR configurations
Consider the syntactic structures in (10).
(10)
a. the ball on the hill

b. The ball rolled down the hill.
English syntax makes it appear that English spatial prepositions describe relations between THINGS ball and hill in (10 a) or between
a processual RELATION and a THING the ball rolling and the hill,
respectively, in (10 b). This is not exactly the case. Consider the semantic
structures in (11).
(11)
a. the road into Saarbrucken

b. the road out of Saarbrucken
c. the road through Saarbrucken
Although I have held constant the THING elaborating the respective

TR valences of into, out of, and through, there is a clear distinction in
the image attributed to this THING the road in each case. The
332
Bruce W. Hawkins
distinction is configurational. When elaborating the TR of into in (11 a),

the road takes on the image of what I call the TERMINATIVE PATH
configuration. When the same nominal elaborates the TR valence of out
of in (11 b), it takes on a rather different image, which I call the INITIATIVE PATH configuration. And when that same nominal elaborates the
TR of through in (11 c), it takes on yet a different image, which I call the
IMPERFECTIVE PATH configuration. The exact same configurational
contrasts are apparent in other sets of English spatial prepositions.
(12)
a. I flew to Chicago.
b. I flew from Chicago.
c. I flew here via Chicago.
(13)
a. The ball rolled onto the newspaper.

b. The ball rolled off the newspaper.
c. The ball rolled across the newspaper.
Common to these three configurations is that they all involve a onedimensional expanse. Any configuration characterized by such a onedimensional expanse is referred to herein as a PATH. In the examples
given in (11), (12), and (13), we encounter three qualitatively different
PATHs. The least complex is the IMPERFECTIVE PATH evident in
(11 c), (12 c), and (13 c), characterization of which requires only what has
already been said: IMPERFECTIVE PATH is purely a one-dimensional
expanse.
Jackendoff (1978, 1983) identifies boundedness as a quality that distinguishes TERMINATIVE PATH and INITIATIVE PATH on the one
hand from IMPERFECTIVE PATH on the other. While any PATH
could conceivably be bounded at two extremes, IMPERFECTIVE PATH
is bounded at neither while both TERMINATIVE PATH and INITIATIVE PATH are bounded at only one extreme. The distinction between
the two bounded PATHs is a matter of which bounding extreme is
included in the configuration. TERMINATIVE PATH includes the extreme I call T E R M I N U S and is not bounded at the opposite extreme,
while INITIATIVE PATH includes the extreme commonly identified as
O R I G I N and is unbounded at the other extreme. Successful characterization of TERMINATIVE PATH and INITIATIVE PATH, therefore,
requires explicit identification of the difference between T E R M I N U S and
ORIGIN. This distinction is a matter of direction, which is ultimately
explainable in terms of the concept of order or sequence. When a PATH
is directed, the continuous set of locations which constitute the PATH
333
must be perceived as occurring in a strategic order. Sequential scanning

(cf. Langacker 1987: 248-253) of such a PATH leads to recognition of
distinct bounding extremes. When sequential scanning results in perception of an ordered change from the one-dimensional expanse of the
PATH to something qualitatively different, that point of change is identifiable as the TERMINUS. When the perceived change is in the opposite
order, from something other than PATH to the one-dimensional expanse
of the PATH, the point of change is the ORIGIN.
In the cases of TERMINATIVE PATH and INITIATIVE PATH, we
find the properties of direction and boundedness co-occurring. However,
not all directed PATHs are also bounded PATHs. Consider the PATH in
the profile of toward in (14).
(14)
The baby is crawling toward the stairway.
This sentence asserts that the infant is headed in the direction of a LM

that can be particularly dangerous for a young child the stairway, but
it doesn't assert that the directed PATH will terminate at that LM. That
is, the DIRECTIVE PATH in the profile of toward leaves open the
possibility that the baby will go right on crawling when she reaches the
stairway, a situation sure to alarm the parent.
Similarly, not all bounded PATHs in English are also directed PATHs.
Consider the semantic contrast between (15 a) and (15 b).
(15)
a. This train runs from downtown Chicago to OHare Airport.

b. This train runs between downtown Chicago and OHare Airport.
Both examples exhibit a PATH bounded at both extremes. I refer to

this configuration as PERFECTIVE PATH. The PERFECTIVE PATH
in (15 a) is the result of explicitly juxtaposing the INITIATIVE PATH of
from with the TERMINATIVE PATH of to. Since both contributing
images are directed PATHs, the composite bounded PATH image is also
a directed PATH. Thus, downtown Chicago is identifiable as the ORIGIN
and OHare Airport as the TERMINUS of this composite PERFECTIVE
PATH.
In contrast, direction is not characteristic of the PERFECTIVE PATH
profiled by between. Although the order of the prepositional objects in
(15 b) may suggest some priority of the first LM, (15 b) is true whether
the train is headed in the direction of downtown Chicago or toward
OHare Airport. Indeed, the best evidence for the truth of (15 b) would
be observations of the same train going in one direction at one time and
in the other direction at a different time.
334
Bruce W. Hawkins
The PATH image evident in (16) is one I call CIRCUITIVE PATH.

(16)
This train goes around the airport once every 30 minutes.
The concepts needed to distinguish CIRCUITIVE PATH from other

PATH configurations are curvature (or, more generally, non-rectilinearity) and closure; CIRCUITIVE PATH is best characterized as a simple
closed curve. The preposition in (17) profiles a TR configuration which
shares the curvature/non-rectilinearity characteristic of CIRCUITIVE
PATH, but not the closure.
(17)
We drove around town for hours looking for our kids.
I refer to this PATH image as NON-RECTILINEAR PATH.

In defining the basic PATH image as one-dimensional expanse, we
have also defined the semantic factor common to directional prepositions.
Every directional preposition exhibits some PATH as its TR configuration. As not all PATH images are directed, we can see that the term
'directional preposition' itself is something of a misnomer.
If the PATH image one-dimensional expanse is the defining
characteristic of directional prepositions, what is the TR configuration
characteristic of locative prepositions? The prepositions we saw earlier in
examples (1) through (6) are all locatives, but those in (18) and (19) are
rather questionable.
(18)
a. There are acorns all over the lawn.

b. There are acorns all across the lawn.
c. There are acorns all around the mighty oak tree.
(19)
a. There is smoke throughout the house.

b. There is smoke all through the house.
c. There is smoke all around the frightened child.
These prepositions are difficult to classify given only a two-way distinction like locative vs. directional. The problem lies in considering the
locative/directional distinction to be a basic dichotomy. These two categories do not exhaust the set of English spatial prepositions. Clearly,
these are the majority categories, but the data in (18) and (19) belong in
a pair of very sparsely populated minority categories. All of these categories are defined explicitly by the nature of the TR configuration in the
relational profile of the preposition. As concluded earlier, directional
prepositions exhibit a profile involving a TR configuration characterized
by one-dimensional expanse. The prepositions in (18) and (19) profile
335
TR configurations which expand beyond a single dimension. The prepositions in (18) exhibit a TR configuration characterizable as two-dimensional expanse, while the prepositions in (19) exhibit a contrasting image
characterizable as three-dimensional expanse. I refer to the former as
AREA and the latter as SPACE.
So what, then, is the configuration characteristic of locative prepositions? Let's put the question in slightly different terms: what configurational information does the image convey? None! That is, locative prepositions profile a relation in which the TR configuration ascribes no
properties whatsoever to the thing or process elaborating the prepositional
TR. Clearly, this doesn't mean that this thing or process has no physical
properties. For this reason, I reject proposals to characterize this configuration as a point an abstract, zero-dimensional entity (e.g., Dirven
1981, Rudzka-Ostyn 1988). I call this configuration NODE, because this
term accurately captures the essence of the TR configuration characteristic of locative prepositions: the conceptual ability to ignore perceived
facts of internal complexity and deal with an entity as an unanalyzed,
autonomous whole a node. This is precisely what cartographers do
when they reduce whole cities to mere dots on a map. Lakoff (1987: 442)
discusses this same basic phenomenon in terms of a multiplex <-> mass
image-schema transformation.
2.2.3. LM configurations
Most analysts of English spatial prepositions recognize the natural classes
in (20), (21), and (22).
(20)
a.
b.
c.
d.
e.
f.
John
John
John
John
John
John
is in the room.
is out of the room.
ran into the room.
ran out of the room.
ran through the room.
scattered papers throughoutjail through the room.
(21)
a.
b.
c.
d.
e.
f.
John
John
John
John
John
John
is on the white carpet.

is off the white carpet.
ran onto the white carpet.
ran off the white carpet.
ran across the white carpet.
scattered papers all across the white carpet.
336
(22)
Bruce W. Hawkins
a.
b.
c.
d.
e.
John
John
John
John
John
is at the front door.

is away from the front door.
ran to the front door.
ran from the front door.
ran (to his car) via the front door.
However, not all analysts recognize that there is a fourth natural class
to be considered along with this group.
(23)
a. You'll find John along the road somewhere.

b. John ran along the road.
c. John scattered papers all along the road.
In comparison with those in (20), (21), and (22), the natural class in
(23) is clearly deficient, but each of these natural classes of prepositions,
including that in (23), is formed on the basis of shared configurational
information about the prepositional LM. There are four LM configurations in English. I call the configuration characteristic of the prepositions
in (20) M E D I U M . The prepositions in (21) share a contrasting LM
configuration most analysts (including myself) call SURFACE. Characteristic of the prepositions in (22) is a configuration we have seen before,
N O D E . And finally, that deficient set of prepositions in (23) share the
LM configuration I refer to as C H A N N E L .
In an earlier forum (Hawkins 1984), I gave characterizations for these
four LM configurations that are functional in nature. Each image was
distinguished on the basis of its relational potential. The relational potential of M E D I U M is characterizable in a number of ways: enclosure,
inclusion, or containment. The relational potential of SURFACE includes
such concrete physical relations as contact, support, resistance, and contiguity. The C H A N N E L image is characterized by a relational potential
that can be characterized as passage, conveyance, or transmission. As
before, N O D E is the odd member of the set in that it seems to have no
specific relational potential.
I refer to this functional characteristic of the LM configurations as
relational potential because the concrete relations actually occur only
when the particular LM configuration is found in a COINCIDENCE
relation. When the same configuration is found in a SEPARATION
relation, the relational potential remains unrealized. For example, locative
out of in (20 b) does not involve a room in any containment relationship.
Instead, it involves SEPARATION from a room which, in some other
situation, could clearly be involved in a containment relation.
337
In a descriptive framework such as this, where configurations clearly

contrast with relations, defining a configuration solely on the basis of its
relational potential seems misguided. Furthermore, reconciliation of the
NODE image with other LM configurations becomes quite difficult, if
not impossible, if the defining characteristic of the set is relational potential. As noted earlier, NODE has no particular relational potential, and,
therefore, would not seem to fit well within a set so defined. Suspecting
that Leech (1969) had come quite close to resolving this problem with
his ascription features, [3DIME], [2DIME], and [1DIME] corresponding to MEDIUM, SURFACE, and NODE, respectively - I confronted
this problem from the same perspective as Leech had (cf. Hawkins 1988).
That is, rather than attempting to confront the MEDIUM image directly,
I followed Leech in recognizing that MEDIUM amounts to the conditions
for acceptable co-occurrence with the prepositions in (20), and that the
nominals which meet these conditions comprise a significant lexical category in English. We can refer to this as the lexical category MEDIUM.
In effect, the MEDIUM configuration is an abbreviated representation
of the lexical category of the same name. If an explicit characterization
can be given to this lexical category, it would reveal the fundamental
characteristics of the MEDIUM image.
My investigation led to the conclusion that the lexical categories MEDIUM, SURFACE, CHANNEL, and NODE are natural categories
defined not by necessary and sufficient conditions but rather by a categorial prototype and local relations of similarity between category members. The condition defining the categorial prototype in each case is a
configurational property: three-dimensional expanse for MEDIUM, twodimensional expanse for SURFACE, and one-dimensional expanse for
CHANNEL. NODE turns out to be a category superordinate to the
others, defined just as we defined it earlier, as an integral, internally
unanalyzed whole. At present, I can see no reason not to equate the LM
configuration in each case with the prototype of the corresponding lexical
category.
3. Implications for a theory of universal grammar

What implications does this particular analysis of English spatial prepositions have for the analysis of other languages? That is, assuming that
the proposed framework has a reasonably high degree of descriptive
338
Bruce W. Hawkins
adequacy for English, what can it reveal about the nature of language in
general? Clearly we cannot claim that the semantic system characteristic
of English spatial prepositions is universal. Indeed, it is safe to assume
that there is probably not another language in the world for which the
proposed descriptive framework would be entirely adequate. So, does the
proposed framework have any utility at all beyond that of describing the
semantics of English spatial prepositions? And if so, what are the implications for a theory of universal grammar?
I have come to the conclusion that this analysis has far-reaching
implications for such a theory. We have seen the semantic system of
English spatial prepositions characterized by three parameters of structure, one relational and two configurational. I suggest that underlying
these particular parameters in this particular semantic domain in this
particular language are two fundamentally distinct, innate parameters of
cognitive structure. One of these is a relational parameter. Analysis of
the spatial prepositions in English reveals that this parameter must include
at least two basic relations: SAME and D I F F E R E N T . Whether this
particular innate parameter includes any other relations is clearly a
relevant question, but not one we can confront productively with the
data presently under consideration. The other parameter involves configurational information. The facts of English spatial prepositions force us
to recognize that this configurational parameter includes at least simple
expanses in one, two and three dimensions and properties which can
make certain expanses more complex boundedness, direction, rectilinearity/curvature and closure as well as the capacity to ignore the
configurational complexities of conceptualized entities.
At this point, it is useful to return to the distinction I drew earlier
between profiled and profilable structures. A profiled structure is a
language-specific semantic structure. It is necessarily highlighted within
some cognitive domain (or set of domains) the base in semantic
structure. Every preposition in English involves a profiled structure which
must meet the following semantic well-formedness conditions: it involves
a basic relation from the inventory in (24) between a TR configuration
from the inventory in (25) and a LM configuration from the inventory
in (26).
(24)
Basic relations
a. COINCIDENCE:
TR and LM occupy the same place in

space, such that there is no perceptible
spatial gap between them.
b. SEPARATION:
(25)
TR and LM occupy different places in

space, such that there is a perceptible
spatial gap between them.
TR configurations
a. NODE:
b. IMPERFECTIVE PATH:
c. INITIATIVE PATH:
d.
e.
f.
g.
h.
i.
j.
(26)
339
internally unanalyzed entity

one-dimensional (1-d) expanse
1-d expanse, directed, and
bounded at ORIGIN
1-d expanse, directed, and
TERMINATIVE PATH:
bounded at TERMINUS
1-d expanse, directed
DIRECTIVE PATH:
1-d expanse, bounded at both
PERFECTIVE PATH:
extremes
1-d expanse, curved/non-rectiCIRCUITIVE PATH:
linear and closed
1-d expanse, curved/non-rectiNON-RECTILINEAR PATH:
linear
AREA:
2-d expanse
SPACE:
3-d expanse
LM configurations
internally unanalyzed entity
a. NODE:
b. CHANNEL:
1-d expanse, relational potential: passage,
conveyance, transmission, etc.
2-d expanse, relational potential: contact,
c. SURFACE:
support, resistance, contiguity, etc.
d. MEDIUM:
3-d expanse, relational potential: enclosure,
inclusion, containment, etc.
In contrast with a profiled structure, a profitable structure is not a

semantic structure; it is a primitive cognitive structure. It differs crucially
from a profiled structure in that the profilable structure is not highlighted
within any cognitive domain. Indeed, a profilable structure exists innately
in the human mind before the process begins of acquiring substantive
information in particular cognitive domains. Throughout this process of
cognitive development, profilable structures facilitate categorization of
incoming perceptual data into packages of information stored economically enough to meet the demands of rapid cognitive processing. We
commonly refer to these packages as concepts. Profilable structures can
be considered innate structural hypotheses which are tested against in-
340
Bruce W. Hawkins
coming perceptual data throughout the process of conceptual development.

My hypothesis about the development of language-particular semantic
systems involves the following three fundamental claims: (i) one or more
innate parameters are implicitly accessed in the search for manageable
semantic structure in a particular system; (ii) some hypotheses from each
parameter are retained for structuring particular concepts within the
system, while others may be dismissed; and (iii) when more than one
parameter is accessed to structure a particular semantic system, the
retained hypotheses in one must coordinate with the retained hypotheses
in the others, resulting in language-specific profiled structures which,
because of this coordination, are clearly more complex than the innate
profitable structures from which they are directly derived.
This hypothesis makes a number of important predictions. First, it
predicts that although distinct semantic domains are structured differently
within a given language, there should be cases of particular abstract
cognitive structures surfacing in semantic structures in a number of
different domains within that language. Second, it predicts that although
a given semantic domain is treated differently across languages, there
should be cases of particular cognitive structures surfacing in the same
domain in different languages. Finally, it also predicts that we could
expect to find instances in which a given innate parameter is accessed
more than once in structuring a given semantic system. In any of these
cases, we should expect to find both similarities and differences in the
profiled structures either across domains or across languages. The similarities would be attributable to the fact that the same resource has been
appealed to in each case, that being a particular parameter of profilable
structures. The differences would be attributable to the fact that this
resource has been applied to substantively different problems in different
contexts.
Thus, we predict both similarity and difference in the manifestation of
profilable structures in three different situations: across domains within
a particular language, in corresponding semantic systems across languages, and within a particular semantic system in a particular language.
It is impractical in the present forum to present and discuss adequately
data relevant to all three sets of predictions. I restrict my attention here
to the predictions which can be illustrated with data from adpositional
semantics. Thus, I will not confront data illustrating the variable manifestation of profilable structures across domains within a given language.
In confronting the issue of consistency and variability across languages
341
a bit later, we will compare the English system of spatial prepositions

with the Japanese system of spatial postpositions. But we do not have to
look to some other language to find our first example of variability in
the manifestation of a given parameter and the structural hypotheses
therein. There is an excellent example within the semantic system of
English spatial prepositions, where the innate configurational parameter
has been accessed not once but twice. As we would predict when the
same resource is applied to two problems, there is significant similarity
across the two subsystems. Both exhibit expanses in one, two, and three
dimensions, as well as the configurationally unanalyzed NODE. We also
predict that in applying a particular resource in distinct contexts, the
contextual distinctions will lead also to some significant dissimilarities
across the two subsystems. This prediction is borne out by the data of
English spatial prepositions in two ways. First, we find that the LM role
has an effect on the three expansive configurations that the TR role does
not. That is, the LM role brings out the relational potential of 1-d, 2-d
and 3-d expanses, while the TR role does not. Second, we find that the
properties of boundedness, direction, curvature/non-rectilinearity, and
closure are coordinated with the 1-d expanse of PATH to yield seven
different TR configurations. In contrast, there is only one 1-d expanse
image among the set of LM configurations: CHANNEL. The configurational property of boundedness does, however, play a minor role within
the set of LM configurations. The LM configuration characteristic of the
two prepositions in (27) is a MEDIUM, i.e., a 3-d expanse, which is
necessarily bounded.
(27)
a. I parked my car inside the garage,

b. I parked my car outside the garage.
Let us turn our attention now to the system of Japanese spatial

postpositions. The first important observation to be made is that Japanese
does not have the rich set of locative adpositional particles that English
does. Instead, Japanese exhibits the constructions no X ni and no X de
along with a rich set of relational nouns to fill the X slot. Given in (28)
and (29) are a few relevant examples.
(28)
a. neko
cat
'The
b. neko
cat
'The
ga
hako no
SUBJ box postp.
cat is on the box.'
ga
hako no
SUBJ box postp.
cat is in the box.'
ue ni
imasu
top postp. be
naka
ni
imasu
interior postp. be
342
(29)
Bruce W. Hawkins
c. neko
cat
'The
d. neko
cat
'The
e. neko
cat
'The
ga
hako no
soto
ni
imasu
SUBJ box postp. exterior postp. be
cat is out of the box.'
ga
hako no
mae ni
imasu
SUBJ box postp. front postp. be
cat is in front of the box.'
ga
hako no
ushiro ni
imasu
SUBJ box postp. back postp. be
cat is behind the box.'
a. neko
cat
'The
b. neko
cat
'The
c. neko
cat
'The
d. neko
cat
'The
e. neko
cat
'The
ga
hako no
ue de
ne
te
imasu
SUBJ box postp. top postp. sleep PROG be
cat is sleeping on the box.'
ga
hako no
naka
de
ne
te
imasu
SUBJ box postp. interior postp. sleep PROG be
cat is sleeping in the box.'
ga
hako no
soto
de
ne
te
imasu
SUBJ box postp. exterior postp. sleep PROG be
cat is sleeping outside of the box.'
ga
hako no
mae de
ne
te
imasu
SUBJ box postp. front postp. sleep PROG be
cat is sleeping in front of the box.'
ga
hako no
ushiro de
ne
te
imasu
SUBJ box postp. back postp. sleep PROG be
cat is sleeping behind the box.'
These data exemplify some important facts about the Japanese system
of spatial postpositions. In order to discuss them with any degree of
adequacy, it is necessary to clarify the semantic composition of the
relevant postpositional structures. The first observation to be made concerns the postposition no. This postposition is used to express a wide
range of relations (cf. Tohsaku 1982), the most common and concrete of
which is possession, as in the NPs in (30).
(30)
a. Yamada-san no
hon
Mr. Yamada postp. book
'Mr. Yamada's book'
b. Watas i
no
kaban
1 singular postp. briefcase
'my briefcase'
343
In the postpositional constructions under consideration, no profiles a

relation more abstract than typical possession. This abstract relationship
involves one nominal, hako, which profiles a discrete bounded entity (cf.
Talmy 1988) and a second nominal ue, naka, soto, mae, or ushiro
which profiles a space contiguous to and, crucially, defined relative to
the first entity. Native speakers of Japanese consistently report that no
does not encode the spatial contiguity relationship between the entity
and this relationally defined space. The reports indicate that the relation
profiled by no is a more abstract one which, I believe, can be identified
as the conceptual dependency of the space upon the concrete object.
This discussion of no leads to an important conclusion concerning the
appropriate parsing of the structures in (28) and (29). The conclusion is
that we are not really dealing with lexically complex postpositions of the
form no X ni and no X de but rather with the simple postpositions ni and
de, the semantic LMs of which are elaborated by NPs of the structure
no N.
This has important repercussions in our analysis of the semantic system
of Japanese spatial postpositions. In effect, we are concluding that the
LM configurational information captured within the English system of
spatial prepositions is not similarly captured within the Japanese postpositional system, but rather, within a separate semantic system of relational nouns. Analysis of this Japanese system of relational nouns could
clearly reveal important facts about the innate configurational parameter
I have hypothesized. For our present purposes, however, the important
fact is that a parameter accessed for a particular structuring task within
the semantic system of English spatial prepositions is not accessed for
the same purpose within the semantic system of Japanese spatial postpositions. This is one scenario which leads to cross-linguistic variability
in structuring a particular semantic system.
Another vital fact that emerges from these data pertains to the inventory of relations profiled by Japanese postpositions. The fact established
just a moment ago concerning the compositionality of the postpositional
LM has a direct bearing here as well. Once we recognize that the
postpositional LMs in (28) and (29) are elaborated not by the simple
nominal hako but by the more complex NP structure hako no X, we must
also recognize that the relation profiled by the postposition ni or de is
that existing not between the cat (neko) and the box {hako), but rather
the relation between the cat and the space profiled by the relational noun.
This relation is consistently COINCIDENCE. In this regard, the contrasts
between the (b) and (c) examples in (28) and (29) take on great significance
344
Bruce W. Hawkins
in that they demonstrate clearly that the notion of SEPARATION is

captured not by Japanese locative postpositions but rather, like the LM
configurational information, within the system of relational nouns in
Japanese. This exhibits another scenario predicted earlier; the same innate
parameter is accessed in structuring both systems of spatial adpositions
English and Japanese. Cross-linguistic variability results from a distinction in the retention and dismissal of innate structural hypotheses.
The English system retains both basic relational hypotheses COINC I D E N C E and SEPARATION. The Japanese postpositional system, on
the other hand, retains only the C O I N C I D E N C E hypothesis. The SEPARATION hypothesis is dismissed in this particular semantic system,
but surfaces elsewhere in the language within the system of relational
nouns.
Finally, we turn our attention to the configurational information Japanese spatial postpositions contain concerning their TRs. Given what we
have already concluded concerning the inventory of relational information and LM configurational information within the system, it should
come as no surprise that TR configurational information is the significant
variable within the system. And, as we would predict, the configurational
structures are not very different from those we find in the English system
of spatial prepositions.
We have already seen that there are two locative postpositions in
Japanese, ni and de. These Japanese locatives are similar to English
locatives in one significant way and different in another. The similarity
is actually a matter of spatial information, the difference a matter of
temporal information. The profiles of both postpositions involve the
configuration I have called N O D E . That is, ni and de share with all
English locative prepositions the fact that the spatial characteristics inherent in the adpositional TR are ignored. Where ni and de differ from
English locatives is that they exhibit a sensitivity to temporal aspectual
characteristics of the verb not exhibited by English locatives. Ni is the
appropriate Japanese postposition to express location when the verb is
stative, while de is the appropriate locative postposition in sentences
involving a dynamic verb.
The directional postpositions in Japanese are (w)o, kara, made and e.
Before discussing these four, however, it is important to note that both
of the locative postpositions occur in sentences involving verbs of locomotion. The role of locative de in such structures is no different from its
role in any other sentence involving a dynamic verb: de simply indicates
where the action is taking place. This is illustrated in (31).
(31)
345
Rooka
de
hasirimasi ta
hallway postp. run
past
was running around in the hallway.'
The role of ni in sentences involving locomotion verbs is rather different. Citing evidence like that in (32), Kuno (1973: 97) suggests that ni
"indicates that the [LM] NP is the goal of the motion designated by the
verb."
(32)
Rooka
ni
hasirimasi ta
hallway postp. run
past
ran into the hallway.'
It is important to note that Kuno's description of the function of ni in

instances like (32) refers explicitly to what I have called the TERMINUS
of a PATH rather than to a TERMINATIVE PATH per se. It would be
inappropriate to conclude from the evidence in (32) that, in structures
involving a verb of locomotion, ni involves the TERMINATIVE PATH
configuration characteristic of the English prepositions to, onto and into.
Such a conclusion would lead one to expect that the structure in (33)
would be interpreted as "I swim to the lake", for example, down a river
to the lake.
(33)
Mizuumi ni
oyogimasu.
lake
postp. swim
swim in the lake.'
But this structure is appropriate only under very special circumstances,

and when it is appropriate it does not involve a TERMINATIVE PATH
in any way. This structure necessitates a more or less permanent locative
relationship between the swimmer and the lake. Thus, this sentence could
be uttered appropriately by a fresh water mermaid describing her regular
activity in her permanent aquatic domain.
A second reason for doubting that ni involves the TERMINATIVE
PATH configuration is illustrated by the data in (34).
(34)
a. Tokyo
Tokyo
went
b. Tokyo
Tokyo
went
ni
ikimasi
postp. go
to Tokyo.'
e
ikimasi
postp. go
to Tokyo.'
ta
past
ta
past
346
Bruce W. Hawkins
c. Tokyo made ikimasi ta

Tokyo postp. go
past
went to Tokyo'.
Although all three of these structures can be translated as went to
Tokyo', they are not synonymous, reflecting the fact that the postpositions
are not synonymous. The only postposition in Japanese with the INITIATIVE PATH TR configuration in its profile is kara, and the postposition used consistently in opposition to kara to indicate a TERMINATIVE PATH is made. This is true whether these postpositions are
used to indicate relations in the spatial domain, as in (35 a), or in the
temporal domain, as in (35 b).
(35)
a. Kyoto kara
Osaka made hasirimasi ta.
Kyoto postp. Osaka postp. run
past
ran from Kyoto to Osaka.'
b. Ku-ji kara ju-ji made hasirimasi ta.
9:00 postp. 10:00 postp. run
past
ran from 9:00 to 10:00.'
The only Japanese postposition to ascribe to its TR all of the configurational information characteristic of TERMINATIVE PATH is made.
The postpositions ni and e are appropriate in contexts involving a PATH
that could be conceptualized as a TERMINATIVE PATH, but neither
conveys all of the configurational information characteristic of TERMINATIVE PATH. As noted earlier, the effect of ni in directional
structures is to highlight some especially significant location. The location
in (33) is significant because of its permanence. Native speakers of
Japanese note that the speaker of (34 a) not only has made a trip that
terminates in Tokyo, but intends to establish residence there. It might be
more accurate, then, to translate (34 a) as 'He went to Tokyo for good.'
Thus, the location indicated by ni (34 a) seems to be significant for two
reasons: (i) it is the TERMINUS of a PATH; and (ii) there is a sense of
permanence with this particular TERMINUS. The profile of ni includes
the NODE TR configuration, and there is information prominent in the
base of semantic structure which makes this NODE identifiable as a
TERMINUS.
I concur with the standard textbook account of e that it highlights the
direction of a PATH. In this regard, it is significant that the train in
(36 a) definitely completes its trip in Tokyo, while that in (36 b) will make
347
a stop in Tokyo and then continue in the same direction to some final
destination beyond Tokyo.
(36)
a. Kono
this
'This
b. Kono
this
'This
kisha wa
train subj.
train goes to
kisha wa
train subj.
train goes to
Tokyo made ikimasu.

Tokyo postp. go
Tokyo.'
Tokyo e
ikimasu.
Tokyo postp. go
Tokyo.'
I conclude, then, that in structures involving a process of locomotion

that reaches some destination, the profile of made includes the TERMINATIVE PATH configuration common to the English prepositions to,
onto and into, the profile of ni includes the NODE configuration against
a semantic background that makes it identifiable as TERMINUS, and
the profile of includes a DIRECTIVE PATH like that we find in the
profile of English toward. Consequently, we find that the profiled structures in these Japanese directionals exhibit no TR configurational information not found also in English spatial prepositions. The profiled
structures may differ across languages, but the same set of profilable
structures can account for the profiled structures in each language.
Finally, we turn to the postposition (w)o in locomotion structures like
those in (37).
(37)
a. Rooka
hasirimasi ta
hallway postp. run
past
ran down/through the hallway.'
b. Miti
arukimasi ta
street postp. walk
past
walked along/down the street.'
Kuno (1973: 97) concludes that (w)o "indicates that the motion designated by the verb takes place covering the entire dimension (or major
portion thereof) of the NP continuously and uni-directionally." In the
terminology of the present analysis, we can say that the profile of (w)o
includes the TR configuration IMPERFECTIVE PATH.
4. Conclusions and inconclusions

The English system of spatial prepositions and the Japanese system of
spatial postpositions differ in a number of easily observed ways. However,
the two systems are not so different that the descriptive framework
348
Bruce W. Hawkins
proposed by Hawkins (1984) is completely inapplicable to the task of

describing Japanese spatial postpositions. What we find in comparing the
two systems is that not all of the descriptive tools needed to describe
English spatial prepositions are needed for the Japanese postpositions.
The Japanese system involves only two parameters of profiled structure
one configurational and one relational while the English system
involves three parameters two configurational and one relational. The
Japanese postpositional system includes no configurational information
concerning the adpositional LM, while the English system does. Within
the relational parameter, the English system involves two basic relations
- C O I N C I D E N C E and SEPARATION, while the Japanese system
involves only the former. Within the TR configurational parameter, the
Japanese system includes N O D E , TERMINATIVE PATH, INITIATIVE
PATH, IMPERFECTIVE PATH, and DIRECTIVE PATH. The English
system includes all of these configurations and a few that the Japanese
system does not seem to exhibit: PERFECTIVE PATH, CIRCUITIVE
PATH, N O N - R E C T I L I N E A R PATH, A R E A and SPACE.
I have suggested that the similarities and differences across the two
languages can be accounted for by hypothesizing two innate parameters
of profitable structures, one relational and one configurational. These
parameters are fundamental resources for linguistic semantic development. Variability in the manifestations of these parameters in particular
domains and in particular languages is hypothesized to result from the
interaction between these innate resources and the particular linguistic
and pragmatic contexts in which they are accessed. These hypotheses
clearly must be tested further, across languages as well as across domains
within a particular language. At this point, however, we can conclude
that the facts of English and Japanese spatial adpositions seem to be
consistent with the predictions of these hypotheses.
References
Dirven, Rene
1981
"Spatial relations in English", in: Gnter Radden and Rene Dirven (eds.)
103-132.
Gove, Philip Babcock (ed.)
1961
Webster's third new international dictionary. Springfield, MA: G. & C. Merriam Company.
Halle, Morris Joan Bresnan George Miller (eds.)
1978
Linguistic theory and psychological reality. Cambridge MA: MIT Press.
349
Hawkins, Bruce W.
1984
The semantics of English spatial prepositions. [San Diego, UCSD doctoral
dissertation.]
1988
"The natural category MEDIUM: An alternative to selection restrictions and
similar constructs", in: Brygida Rudzka-Ostyn (ed.), 231-270.
Jackendoff, Ray
1978
"Grammar as evidence for conceptual structure", in: Morris Halle Joan
Bresnan George Miller (eds.), 201-228.
1983
Semantics and cognition. Cambridge, MA.: MIT Press.
Johnson, Mark
1987
The body in the mind: The bodily basis of meaning, imagination, and reason.
Kuno, Susumu
1973
The structure of the Japanese language. Cambridge, MA.: MIT Press.
Lakoff, George
1987
Women, fire and dangerous things. Chicago: University of Chicago Press.
1987
Foundations of cognitive grammar 1: Theoretical prerequisites. Stanford: Stanford University Press.
Leech, Geoffrey
1969
Towards a semantic description of English. Bloomington: Indiana University
Press.
Radden, Gnter Rene Dirven (eds.)
1981
Kasusgrammatik und Fremdsprachendidaktik (Anglistik und Englischunterricht
14). Trier: Wissenschaftlicher Verlag.
Rudzka-Ostyn, Brygida
1988
"Semantic extensions into the domain of verbal communication". Brygida
Rudzka-Ostyn (ed.), 506-553.
Rudzka-Ostyn, Brygida (ed.)
1988
Topics in cognitive linguistics. Amsterdam: John Benjamins.
Talmy, Leonard
1988
"The relation of grammar to cognition", in: Brygida Rudzka-Ostyn (ed.),
165-205.
Tohsaku, Yasu-hiko
1982
Japanese: language and culture. San Diego: UCSD.
Interpreting and translating prepositions:

A cognitively based formalization 1
Cornelia Zelinsky-
Wibbelt
1. Basic assumptions
In this paper we wish to present a topic which deals with a general
phenomenon of translation: the fact that the same real situation is often
expressed differently in different languages. This has always been one of
the main problems of human translation. The motivations for different
usage patterns in a foreign language are often only learned by long
experience of the cultural environment of that language. This means that
speakers should learn a foreign language in much the same way as their
mother tongue: proceeding from an inborn cognitive disposition, the
language learner might elaborate his linguistic competence in accordance
with the conventional setting of the specific foreign environment. Seen
in this way, translation from one language into another is basically a
relation between different elaborations of the same cognitive endowment.
The aim of this paper is to investigate this translation relation in terms
of the cognitive process by which it is accomplished, and to formalize
this process. Whereas many linguists deal with an operationalization and
formalization of the mental processes underlying language performance,
we pursue these aims in our specific bi-lingual environment. Thus our
formalization may function as a model for an implementation in a
machine translation (MT) system. The model has already proved successful and its implementation is exemplified e. g., in Zelinsky-Wibbelt
(1990 b, 1991 and 1993 forthcoming).
At the first glance it might appear far-fetched to many readers with
regard to the topic of this paper to relate mental processing to natural
language processing and ultimately to machine translation. We hope,
however, that in developing our ideas throughout this paper it will become
clear that it is exactly the relation between different exploitations of some
common linguistic disposition in different social settings which has to be
grasped in a formal mapping between different languages. It is, however,
not within the scope of this paper to give an exhaustive model of the
mental process of translation. We are rather looking for a minimal set
352
of cognitive principles and rules by which speakers are constrained in

interpreting and producing utterances and in linking the interpretation
of the source language utterance to the production of the target language
equivalent.
Thus it is our assumption that different cognitive elaborations effected
by different environments result in different usage patterns. We hold that
different usage patterns are due to different conceptualizations of the
same real situation. This constitutes a fundamental claim of cognitive
semantics:
An expression's meaning cannot be reduced to an objective characterization of the situation described.
Taking this claim seriously has important consequences for the semantic
interpretation of an utterance. The interpretation of an utterance has to
be approached from two points of view. On the one hand the mere
propositional content of an utterance is interpreted, while on the other
hand the meaning of specific usage patterns is considered. This dichotomy
acquires particular relevance from our contrastive point of view. Translation always has to deal with the two parameters of sameness and
difference: on the one hand the reliance of language on canonical conditions, which constrain the interactions of human beings, results in crosslanguage similarity to a considerable degree. The most outstanding constraint which has cross-cultural validity is the gravitational system. On
the other hand the interaction of human beings is determined to a high
degree by their social environment, thus resulting in cross-language diversity. This means that the interpretation of an utterance has to consider
beyond the purely interlingual patterns of meaning the languagespecific usage patterns of an utterance. These in turn can only be given
their specific meaning if we investigate the conditions or motivations
which have given rise to them.
The bi-lingual relations of sameness and difference demand a linguistic
specification of the conditions under which utterances are encoded and
decoded in different languages. Cognitive Grammar holds that the following restrictions constrain linguistic expressions in general (cf. Langacker 1987: 99 ff.):
the background assumptions and expectations of speaker and hearer

the perspective from which the speaker views the situation
the degree of specificity to which the situation is conceived
the salience and relevance attributed to the parts constituting the
conceived situation
Interpreting and translating prepositions
353
These are conditions which have traditionally been considered as "nonlinguistic" or "pragmatic" and have been isolated from an "autonomous"
linguistic investigation. Recently these cognitive constraints have been
considered for word sense disambiguation and for the coherent semantic
interpretation of sentences as well as for lexical acquisition in different
natural language processing (NLP) systems (cf. Dahlgren 1988; Dahlgren
- McDowell - Stabler 1989; Hobbs et al. 1987; Moens et al. 1989;
Miller Johnson-Laird 1976; Annick Pustejovsky 1990; Boguraev
Pustejovsky 1990; Briscoe Boguraev 1990; Pustejovsky 1989; Wilensky
1990). As these cognitive constraints are exploited differently depending
on the specific environment of a language, they are crucially important
for the translation of one language into another.
For our specific purpose of machine translation this insight means that
we need a conceptual motivation for different usage patterns in different
languages. Beyond the interpretation of the mere propositional content
of an utterance, which would result in an interlingual representation of
meaning, we have to interpret how the conceptualizer chooses to construe
the situation in order to express it accordingly (cf. Langacker 1987).
Otherwise machine translation will be massively underdetermined, and
the quality of the translation will be disastrous. Since, according to their
pragmatically determined exploitation, the cognitive constraints result in
a language-specific elaboration, an interlingua in its strict sense (cf., e. g.,
Nirenburg Raskin Tucker 1985) is not sufficient for relating translational equivalents of different languages. The pragmatically based language-specific elaborations also have to be represented and related, in
order to generate well-formed surface expressions in a satisfactory way.
It is our aim, however, not to relate these language-specific elaborations
directly in a transfer component. The ideal is rather to derive an interlingual abstraction, established by the intersection relating the conceptual
representation of the source language to that of the target language. The
language-specific elaborations will only be used in analysis and synthesis. 2
This reduces our interest in cross-linguistic diversity to two main points:
(1) Which are the translationally constant properties which constitute
the "tertium comparations" necessary for mapping expressions from
one language into another?
(2) Which are the language-specific properties which are necessary for
an exhaustive monolingual analysis and synthesis in order to yield
semantic representations from which well-formed surface expressions
may be generated?
354
Irrespective of whether they have cross-linguistic or language-specific

significance we consider the cognitive constraints to be conceptual motivations of meaning in the broadest sense, namely in the sense that
abstract situations, which may not be perceived sensorially are conceived
in terms of concrete, sensorially perceivable scenes. This implies a farreaching assumption:
Conceptualization is basically a process determined by perceptual
principles, (cf. Lakoff Johnson 1980).
We will refer to this process as perceptually driven conceptualization.
However, our main concern in this paper will be the interpretation and
translation of semantically concrete expressions, which are directly driven
by senorial perception. The pervasive influence of sensorial perception
on abstract domains will only be touched on.
2. Area of investigation
In this paper we aim to illustrate our basic assumptions with a special
type of relational expression, i.e., prepositions. Prepositions rank among
the most highly polysemous expressions, and in the last decade interest
in polysemy and in prepositions in particular has grown as a consequence
of the Wittgenstein rebirth, which has ocurred in the wake of the research
paradigms of Lakoff and Langacker. One reason for the prevailing
interest in prepositions is that they establish a nearly closed class and
that their relational meaning consists in a small number of very abstract,
i. e., simplified geometric and topological properties. Therefore they seem
relatively easy to control, although the abstraction of these properties is
a complex cognitive process. On the other hand prepositions relate at
least two arguments containing open-class, so-called "referential" expressions. The term "open-class" means that they undergo continuous
lexical change and development of novel word senses and word forms.
Thus there may be a steady interference between the lexical change of
the referential expressions and the prepositions' semantic constraints on
these. Therefore it is not only on account of their rare but still observable
participation in word-formation, but also because of their rather strong,
yet not obligatory restraint from semantic development, that the position
of prepositions as lexical categories lies in between "closed" and "open"
class categories, i. e., they are intermediate with respect to lexical change.
355
For this reason they are particularly suited for investigating general
phenomena which are lexically driven, namely lexical vagueness, ambiguity and flexibility, i. e., lexicalization patterns involving the development
of novel senses of a lexical unit, in short, the organization of the lexicon
as well as its integration within language as a whole. Seen in this light
the semantic interpretation and translation of prepositions raises several
interrelated problems, the solution of which will offer many important
insights of a more general status which may then be easily adapted to
other lexical categories:
(1) In which way are the syntactic and semantic functions of prepositions
related?
(2) Which information is basic to a preposition and which information
is evoked, specialized or constrained by a specific use in a specific
context? This concerns both the granularity, i.e., the specialization
of a dimension of meaning, and the number of dimensions, which
together determine the regular usages of a preposition.
(3) Which are the criteria for distinguishing between different senses of
a preposition? To which degree are these distinctions conventional,
generic and intrinsic? Langacker's criterion for conventionalized
meaning is the regular use of the respective linguistic unit by a
significant speech community (cf. Langacker 1987: 159 ff.). Generic
meaning comes about through a high frequency relation between a
semantic category as type and its corresponding exemplars as tokens.
This relation becomes increasingly more intrinsic, the less context is
necessary for its instantiation.
(4) How does one account for the relation between physical source senses
and abstract target senses of prepositions? The mental process of
metaphorical derivation becomes especially relevant in the computational processing of linguistic meaning, where novel uses can be
interpreted by analogical reasoning, thus operating on regular relations between concrete and abstract domains.
(5) Considering the preceding points, which is the most natural and most
efficient representation and processing of the meaning of prepositions?
(6) Which parts of a situation are within the range of a prepositional
predication?
356
3.
Word meaning and context
3.1. The holistic view

Each of the questions we have raised in Section 2. as to the investigation
of prepositions is related to the fundemantal question of how prepositions
should be represented in the lexicon with respect to the operations which
work on their representations. Therefore in this section we outline our
general notion of a semantic lexicon within the language system.
We consider natural language to be part of the overall cognitive
capacity of human beings and thus to establish an integrated system in
which various competences interrelate in a complex way. Although such
competences as syntax, semantics and phonology are qualitatively different, one cannot function without the other: these language components
establish a structural continuum with qualitative differences between
those parts in which the structures of different faculties are realized.
Multi-word units are a clear example of the continuous transition between
grammar and the lexicon. This holistic view prevents one from viewing
linguistic components, units, categories etc. as autonomous and studying
them in isolation without any possibility of considering constraints existing between them.
It is because of the continuous nature of language that every division
into discrete linguistic units, such as components and categories, is made
for reasons of linguistic analyzability. It is for the same reason that
speakers mentally classify their enormously heterogeneous environment
into homogeneous semantic categories: they do this in order to reduce
the vast amount of information provided by reality, so that it becomes
mentally processible and storable. Thus it does not follow from the
continuous nature of language that the distinction into units is not
justified. On the contrary: mental categorization as a means of economical
information processing permits linguists to abstract discrete linguistic
units and components, as long as their model of representation and
processing considers the continuous transition between units. By this
means of linguistic categorization we will certainly gain deeper insights
into mental categorization.
Mental categorization, however, is not only a means of economical,
but also of flexible information processing: by virtue of the plasticity of
semantic categories speakers may steadily make use of creative categorization by unifying novel tokens with old types (cf. Jackendoff 1987:
137), thus adapting language to changes in their environment. Despite
357
this continuous dynamism, economy is preserved, since ever more instances are related to the same category. It is this flexible use of knowledge
which spings from meanings that are vague as to their borderlines and
components that has to be foreseen by a model of computational processing. All this means that the linguist must be aware of the parameters
effecting the qualitative differences between categories, units, components
etc. on the one hand and of those parameters by which different components are interrelated on the other hand.
Considering the particular purpose which an N L P system should obey,
it becomes all the more clear that the lexicon is both a theoretically and
empirically as well as a practically guided abstraction. As such the
semantic lexicon encodes a central part of our knowledge about the world
and relates to all knowledge types. Lexicalization depends on the degree
of conventionality, genericity, intrinsicness and typicality which a particular speech community attributes to the knowledge which is associated
with a word at a particular point in time (cf. Langacker 1987: 159 ff.).
Langacker (cf., e.g., 1987: 147ff.) has introduced the notion of basic
cognitive domains for those meaningful properties of a word, which are
mentally irreducible and on which knowledge of any complexity is based.
Systems in artificial intelligence often encode knowledge about specific
domains, which, depending on the domain described, represent word
meanings differently, as their typical properties and functions vary from
one domain to another. This domain-specific knowledge is closely interrelated with pragmatic knowledge about how language is used in certain
utterance situations, in certain text types etc. Langacker does not make
the traditional distinction between semantics and pragmatics or between
competence and performance. Rather, all knowledge types constitute the
conceptual encyclopedia. Linguistic meaning arises from all knowledge
sources which may be accessed by the speakers of a speech community
at a particular point in time; thus utterances obtain meaning as part of
some encompassing cognitive routine, which mirrors the speaker's interaction with reality. What follows from this assumption, is a broad notion
of compositionality as well as a broad notion of context, which by
consideration of all types of encyclopedic knowledge and of all kinds of
the speaker's environment allow new word meanings to be generated in
situations in analogy to productive lexical patterns. In a very broad sense
the meaning of a word then is "the pattern of affinities and disaffinities
with all other words in the language with which it is capable of contrasting
semantic relations in grammatical contexts" (Cruse 1986: 16). This is a
definition of word meaning which is basically relational. The pattern of
358
word meanings has to be explored by finding generalizations, redundancies and defaults of the semantic relations corresponding to the grammatical contexts in which words are used. Generalizations are schematic,
i.e., they are more general than their contextually specialized instantiations. Redundancies are recurrent patterns of meaning, which are highly
productive and from which novel uses of a word may be derived. The
default meaning is that sense which, from a graded structure of possible
word-senses, is instantiated as the typical meaning in a normal context.
With this organization of lexical knowledge in mind linguists will
operationalize the abstraction of basic patterns of meaning, holding
universally across domains and utterance situations. These basic semantic
patterns will be achieved by analyzing different sources of knowledge:
knowledge about specific domains, knowledge about language use, knowledge achieved through general behaviour etc.
3.2. Distinguishing between different senses of a preposition

Word meanings compose the meaning of larger semantic structures in
various ways, according to which the adequate word sense or lexical unit
has to be selected during the generation of a semantic representation.
Herskovits claims that word meanings in general can only be defined by
their compositional properties, i. e., by exploring how they take part in
the composition of larger regular semantic structures, although, in line
with prototype semantics, she admits that Frege's principle cannot fully
explain the creative use of language. Her investigations into the meanings
of prepositions are undertaken from a broad perspective in order to come
up with clues as to the meanings of words in general. With this aim one
can distinguish the meanings of different prepositions as well as different
senses of a preposition by their constraints on the spatial configurations
they designate. Herskovits distinguishes between constraints bearing on
the participants of a spatial scene directly and constraints bearing on the
context of discourse (cf. Herskovits 1986: 919). Among constraints bearing on the scene directly she distinguishes:
(1) Selectional restrictions holding between the prepositional predicate
and its arguments. These are the interdependencies holding between
the relational concept expressed by the preposition and the arguments
which it relates. For instance the German preposition seit has two
359
senses according to its selectional restrictions and these senses surface

in different prepositions in English as the following example shows:
(1) Wir warten seit drei Uhr => We have been waiting since three
'clock
[ TEMPORAL EXTENSION]
(2) Wir warten seit drei Stunden => We have been waiting for
hours
three
[ + T E M P O R A L EXTENSION]
The first sense of seit 'since' relates the beginning of a TEMPORAL

EXTENSION with a POINT IN TIME, that is, it requires its second argument
to refer to a point in time. In its second sense seit 'for' expresses the
relation between a TEMPORALLY EXTENDED activity and the period of
time in which this is temporally located, requiring its second argument
to refer to a period of time.
In examples (1) and (2) the preposition's meaning could be identified
by means of the default meaning of the N P arguments. In the next
example the selectional restrictions can only unify with the N P argument by considering specific background knowledge:
(3) Seit der Wiederaufbauphase
bilindustrie entwickelt.
nach dem Krieg hat sich die Automo-
Since the postwar reconstruction period the car industry has been
developing.
?For the postwar reconstruction period ...3
In this example the noun Wiederaufbauphase clearly does not refer
to a period of time, which would be assumed as its default, if a
context like this were missing. Rather, by considering historical
knowledge about the temporal location of events, Wiederaufbauphase
is associated with an event located in the past and because of its
termination is either projected to the beginning or ending point of
this event or in case of vagueness of the starting point of the development of car industry within the postwar reconstruction period the
whole period is projected to a point. This explains the use of since in
English, which mirrors, that the preposition relates a period to a
POINT IN TIME.
(2) Possible spatial relations holding between the related entities. In a

specific spatial relation the conceptualization of the second argument
360
may constrain the conceptualization of the first argument. The semantic inacceptability of ?the tree standing along the river results from
the mismatch of river and tree in this relation: the preposition along
forces river to be conceptualized as a linear PATH with an EDGE parallel
to which a one-dimensional entity must be conceptually located. This
is not possible with tree in contrast to its plural form trees in the
trees standing along the river which we idealize to the one-dimensional
line connecting the individual trees.
(3) The relational concept of a preposition may constrain the indexical
meaning of trajector and landmark. These are conditions on trajector
and landmark which by an implied norm or normality or a contextually rendered orientation resolve vague relational concepts. It may,
for example, depend on the proportion between trajector and landmark whether prepositions such as between, right of, in the corner are
true or not.
(4) Constraints on "metonymic" conceptualizations of objects. This is
what Herskovits designates as geometric description functions (cf.
Herskovits 1986: 57) which constrain our conceptualization of some
real world entity in terms of different geometric idealizations. We
may for instance conceptualize only the SURFACE of a cube or some
EDGE of it as its salient part in a given context. The above given
example ?the tree standing along the river shows, that the tranformation into different geometric entities is not arbitrary. The meaning
of tree may not be extended to refer to a straight line, hence the
semantic anomaly.
Among constraints bearing on the discourse context Herskovits distinguishes the following pragmatic circumstances:
(1) Constraints on the entity's location by the purpose of the relational
expression. By uttering the sentence The children are going by bus a
speaker will probably answer a question like How are the children
getting there? In contrast to this the sentence The children are riding
on the bus will answer a question like Where are the children now?
(2) Constraints on highlighting some background element associated with
some part of the spatial scene. Thus the sentence There is a gas station
on the road to London will be uttered in a situation in which the event
of someone travelling along the road is conceptualized, CONTIGUITY
between the boundaries of the gas station and the road is thereby
expressed. In contrast to this the sentence There is a gas station at
the road to London only indicates the COINCIDENCE of the position of
361
the gas station with that of the road, at instantiating the projection
of the gas station to a ZERO-DIMENSIONAL point and that of the road
to a ONE-DIMENSIONAL line respectively. This sentence will not necessarily be uttered with a travelling event in mind, that is the speaker
is not mentally near to the spatial scene.
3.3. Lexicalization patterns

It has always been a main concern of lexicology to draw the line between
lexical homonymy, polysemy and lexical vagueness. This is the distinction
between two types of differing usage patterns of a word: the context may
instantiate the intended sense of a word in different ways. Modification
may consist in highlighting different parts, aspects or attributes of the
respective entity. This process is also called modulation, because the same
entity is referred to in each case (cf. Cruse 1986; Taylor 1989: 124);
different highlightings of an entity are not necessarily in disjunctive
relationship. The context modulates lexically vague dispositions of one
and the same sense in different ways.
Modification as the selection or generation of one of several polysemous, disjunctively related senses of a lexical unit consists in the exclusion
of the other senses, i.e., a polysemy pattern subsumes several mutually
exclusive, but related senses, the use of which depends on the respective
discourse situation.
Theoretically the differentiation of the senses of a word depends on
what one assumes to be part of lexical information, i. e., which properties
one assumes to be inherently or intrinsically associated with entities in
general. The distinction between inherent and non-inherent information
is not a strict one in the sense of analytical truth conditions, as, in
accordance with cognitive principles, it may be possible to override
inherent properties in highly unusual situations. Pragmatically the distinction between what is lexically polysemous and what is lexically vague
depends on the conditions imposed on the uses of words in possible
discourse situations: certain properties may remain vague or may be
modulated, certain properties have to be obligatorily selected both in
encoding and decoding utterances.
In the cognitive paradigm prepositions are generally considered ideal
for the study of lexicalization patterns for the following reasons:
(1) prepositions establish a nearly closed class of lexical units, the meaning of which is mentally represented in terms of highly abstract, i.e.,
simplified properties;
362
(2) prepositions originally are spatial predicates and concrete space is

better accessible than its abstract derivate;
(3) according to the strong hypothesis we defend, all semantic extension
at some time has originated from spatial senses and therefore the study
of our conceptual means to structure concrete space at the same time
supplies us with clues for our mental organization of semantic space in
general (cf. Talmy 1983: 226); for instance the distinction into true lexical
homonymy, polysemy and vagueness by the cognitive principles driving
these meaning relations may be obtained in a relatively clear way with
respect to the physical domain;
(4) prepositions are highly polysemous with dense patterns of metonymy
and metaphor.
To a certain degree polysemy patterns are highly productive and the
issue is currently being discussed in computational linguistics under the
notions of economy and flexibility. This discussion raises the question of
whether productivity is high enough to result in significant regularities,
which may be accounted for in more general terms than in recurrent
identical descriptions, by which the lexicon becomes highly redundant.
Different options of representing and processing polysemy patterns are
suggested:
The schema model claims that lexical information should be as general
as possible in abstracting from different contextually instantiated sense
specializations. Sowa (1984); Bierwisch Lang (1989); and Habel
Herweg Rehkmper (1989) defend different versions of this idea. It is
common to all versions of the schema model that a strict distinction is
made between linguistic-semantic knowledge encoded in the lexicon and
conceptual knowledge induced by the context.
The prototype model assumes different polysemes of a word to establish
a family resemblance structure, which may be graded in that there is a
typical, most representative sense of a word, to which continuously less
typical senses are related; Barsalou (1992) assumes that typicality derives
from frequency of instantiation. Boguraev and Pustejovsky (1990) assume
logical metonymy relations between senses and logically derive metonymies and metaphors from lexically encoded defaults; in the absence of
contextual restrictions this may be taken to establish the typical, most
representative sense.
The commonsense model proceeds from Putnam's (1975) idea of stereotypical knowledge about words. Dahlgren (1988), Hobbs (Hobbs et al.
1987) are defendants of this model. They claim to represent the average
everyday knowledge of speakers. In the same way as prototpye semantics
they consider semantics to be encyclopedic. A major concern of naive
363
semantics is the relation between knowledge about typical properties in

specific domains and kind types, which establish very basic classificatory
knowledge. Dahlgren assumes a single entry for each words sense, as the
typicality patterns of word meanings are domain-dependent, and kind
types constrain these in specific ways for each sense.
The sense generation model of situation semantics claims different senses
of a word to be anchored in real situations and to be generated nonmonotonically in accordance with the respective discourse situation (cf.
Franks - Braisby 1990; Myers - Franks - Braisby 1989). The default
meaning corresponds to the contextfree lexicalized meaning (cf. Myers
1989: 8). In principle, there is no constraint on defeating a property
associated with a word.
Cognitive Grammar refrains from accepting any of these suggestions
as an exclusive model of language. It rather seems that several types of
lexical abstraction have to be considered by a model of the mental lexicon.
Both schemata and prototypes are economical mental strategies for abstracting from the vast amount of information offered by reality. The
two strategies work in close interrelation. For instance, the exclusive use
of schemata has often proved to be insufficient, as schemata are too
coarse and not flexible enough for drawing inferences. The reason for
this is that similarity relations between polysemes are not necessarily
transitive and that one schematic abstraction does not necessarily generalize over all senses of a polysemy pattern of a word. Cognitive Grammarians also emphasize that metonymy relations are not completely
regular. Lehrer (1990: 240) concludes on an empirical basis that, although
there are many miniregularities, only a few rules are without any exceptions. This implies that the lexicon must also contain redundant information. Although these irregularities may be accounted for by nonmonotonic procedures, the latter are constrained by cognitive priciples,
that is, contrary to situation semantics we hold that in a specific pragmatic
environment certain semantic properties may not be defeated.
It follows that the relations between polysemes determine the degree
of lexical redundancy, which has to be somewhere between a pure enumeration of the senses of a word and the association of a word with a
regular polysemy pattern as a semantic type.
With prepositions the interaction between different types of lexical
abstraction is assumed to be the following: a preposition may express a
family of senses, the relations between which consist in a generalized
schema and an ideal. In contrast to a prototype, which represents the
best exemplar of a category by characterizing it in fine detail, an ideal
establishes an idealization of typical properties, in terms of a reduction
364
or simplification. Ideals are repesented in terms of relatively simple

structures, compared to our bodily experience they are derived from. As
such, Lakoff calls them image schemata (cf. Lakoff 1987: 267). It is
common to all types of lexical abstraction that all senses may be related
to it, either by way of specializing some abstract schema, or by different
transformations, which relate the ideal meaning to all its metonymical
and metaphorical derivations radiating from it in the way modelled by
Dirven in this volume. One purpose of modelling polysemy in this way
is the interpretation of novel uses (cf. Herskovits 1986: 40).
For instance, the preposition in has one ideal meaning, it expresses the
relation of INCLUSION, requiring its second argument to refer to a COMPLETELY BOUNDED THREE-DIMENSIONAL entity. Around this, all derived
senses center by exploiting and transforming the ideal meaning in different
ways. Whereas the PP the fly in the box expresses an ideal inclusion, in
the PP Big Ben in London the preposition in is semantically related to
the prototypical sense relation: London is idealized to a TWO-DIMENSIONAL
AREA and thus can only provide a PARTIAL ENCLOSURE. Yet the relation
of INCLUSION holds, although in a less typical way. In the same way as a
prototype the ideal meaning does not establish a common core of the
category. This may be illustrated with the preposition on. With the lexical
pattern of this preposition the notion of a family of senses, which
Wittgenstein has set against the core meaning notion, becomes obvious:
in its ideal meaning the preposition on seems to express the idea of
SUPPORT which implies CONTIGUITY of the parts of the scene related so
that the larger part of the scene provides its SURFACE on which the other
part is located and thus supported. This ideal meaning is expressed by a
vase on the table. There is, however, a meaning of on which lacks the idea
of SUPPORT, only implying CONTIGUITY. This concept may be considered
to be derived from the ideal meaning. Again this derived concept is
exploited in several different ways. Table 1 illustrates the concept of
CONTIGUITY, in which the senses are given in order of decreasing typicality
(cf. Herskovits 1986: 40 ff.).
Table 1. The concept of CONTIGUITY

EXAMPLE
CATEGORY
Two physically extended entities in
CONTIGUITY
relation
a lock on the forehead
Physically extended entity projected to be C O N T I G U O U S with

the E D G E of a T W O - D I M E N S I O N A L L Y extended entity
a house on the park
Visually apparent
shadows on the wall
CONTIGUITY
between two entities pro-
j e c t e d t o TWO-DIMENSIONAL AREAS
365
4. What do prepositions express?

Within the framework of Cognitive Grammar an old assumption of
formal semantics has undergone a revival: prepositions are considered as
two- or three-place predicates, because they express a relation between
two or three speech participants. Hence Langacker classifies prepositions
as relational expressions, because they express how the conceptualizer
configurates the parts which constitute a spatial scene with respect to
each other. Thus prepositions express how the speaker establishes a
cognitive link between two or three parts of a spatial scene. This cognitive
relation is profiled in accordance with cognitive principles operating on
the conventions associated with the respective parts of a spatial scene in
the particular cognitive domain. These conventions yield one part as the
new part which is standing out as the foreground from an already known
part establishing the background, which functions as a location. In this
partitioning the foregrounded figure receives prominence, because in a
particular cognitive domain it is mobile or more movable in relation to
its location which usually is more stationary. Langacker (cf. e.g., 1987:
217 ff.) has introduced the terms of trajector and landmark for these old
gestalt-psychological distinctions in order to mark their linguistic function. The foreground is called the trajector and the background the
landmark. Both are constrained with respect to the conceptualizations
they can undergo (cf. Hawkins this volume). The landmark functions as
a reference point with respect to which the trajector's motion, location
or orientation is specified. In European languages this location is typically
designated by a nounphrase in object position. The trajector may be
either an entity designated by a subject N P or a relation, which is typically
designated by a verb of motion, position or direction in the spatial
domain:
(4)
(5)
The cat is on the mat.

The cat is playing on the mat.
Sentence (4) expresses the spatial relation between two entities, between
the cat as the trajector, and the mat as the landmark. Sentence (5) relates
the cat's activity of playing, the trajector, to the mat's location, the
landmark.
These examples show that the conceptual partitioning of the real spatial
scene is asymmetrical. We can say that under normal pragmatic conditions
the cat in sentence (4) is the more movable entity, as it is MOBILE. Therefore
it is focussed as standing out in the profiled relation and hence functions
366
as the trajector. In sentence (5) the same holds for the cat's activity of
playing. In both sentences the mat as the IMMOBILE entity is less movable
and therefore the less prominent backgrounded entity with respect to
which the trajector is located. The asymmetrical relation between trajector
and landmark becomes most obvious if we try to turn the relationship
around as in the following examples:
(6)
(60
The cat is on the mat.

?The mat is under the cat.
(7)
(70
The lid is on the teapot.

?The teapot is under the lid.
(8)
(80
Marseille lies on the Mediterranean.

*The Mediterranean lies on Marseille.
A speaker would only utter sentence (6'), if, contrary to the normal
situation, his focus of attention did not rest on his cat, but on the correct
position of the mat. (7') would only be acceptable in an extremely unusual
situation in which normal pragmatic conditions do not hold. The normal
function of the lid is to cover the teapot, which would be expressed by
the preposition on. Without any context from which these conditions
may be inferred a normal hearer's assumptions are violated by this
sentence. For sentence (8') there seem to be no exceptional conditions at
all under which it would be semantically acceptable. Although the Mediterranean may doubtlessly be conceived as lying between Europe and
Africa, its extension does not allow for the conception that it lies on
Marseille, since the use of the preposition on has to follow the condition
that the trajector is smaller than the landmark.
Prepositions may basically express two complementary relations: COINCIDENCE and SEPARATION (cf. Langacker 1987: 223 ff.; see also Hawkins,
this volume). That is, they express whether some part of the space which
trajector and landmark occupy is the same or whether they occupy
different spatially discontinuous places. Set-theoretically these basic relations may be expressed in the following way:
COINCIDENCE: PLACE ( T R )
SEPARATION: PLACE ( T R )
PLACE ( L M )
PLACE ( L M ) =
These basic relations are logically disjunct, that is, only one of both may
apply. They do not depend on the conventional setting of the environment, although they are related to it insofar as the two basic relations
are assumed to be generalizations over all language-specific ones. That
367
is, the basic relations hold universally across domains and utterance
situations, while the language-specific specilizations centre around them.
Thus the basic relations constitute the tertium comparationis which may
be used for relating the semantic representations of different languages.
5.
Schematization as the process of conceptualizing

spatial relations
5.1. The constitution of schematization

The main point we want to illustrate with the translation of prepositions
is how certain circumstances and purposes of an utterance instantiate
important principles of conceptualization. We will particularly consider
pragmatic factors resulting from the specific environment of the language
user and the utterance situation; we will investigate how these pragmatic
factors determine the relevance, salience and typicality of the entities
constituting the conceived situation. Moreover, we consider these pragmatic factors, which depend on the culture-specific environment of the
language user, to be conceptual motivations of meaning which are perceptually driven, i.e., abstract mental space configurations have originated from physical space configurations.
Being conceptually mediated, reality is not expressed directly by language. Rather, language expresses how speakers have mentally implemented the common experience of the specific environment they live in.
This implies that the semantic distinctions conforming to our linguistic
utterances about our spatial environment do not agree with the real
physical extension of the configurated parts, but with their conceptual
SCHEMATIZATION.
SCHEMATIZATION is the fundamental principle underlying the linguistic
expressions of spatial configurations (cf. Talmy 1983: 225): it is the
selection of those spatial properties of a scene which in a given domain
and situation are essential to how we view it as a whole, while the
remaining properties, being non-essential, do not participate in this process. This selection process relies predominantly on geometrical and topological properties. It operates on two preconditions: While prepositions
as relational predications are highly polysemous with respect to the
possible relations they can express, the entities which they relate are
368
lexically vague with respect to the possible spatial properties they may
adopt in the respective configurations. Each schematization process results in the conceptualization of a schematization type. We will assume
that there exists a limited set of schematization types; these include all
and only the properties essential for us in the conceptualization of space.
The process of schematization may be characterized in terms of the
following complementary strategies (cf. Talmy 1983: 225):
(1)
IDEALIZATION
This is the process by which we make entities which in physical reality

are highly complex and variegated conform to some associated mental
scheme. This process is comparable to the process of prototyping our
environment (cf. Rosch 1978) with the difference that, whereas prototypes, as best exemplars of a category, may be mentally represented in
fine detail, the representation of ideal meanings is reduced to very abstract
geometrical and topological properties. A typical case of idealization is
the projection of a three-dimensionally extended object to a one-dimensional line when its designating noun precedes prepositions such as
English along as in the trees along the river. Here the trees, which in
reality are dicontinuously located, are idealized to a continuously extending one-dimensional line, as required by the ideal meaning relation
expressed by along.
Thus the boundary conditions of an entity, which constitute the enclosure, may be idealized by what Hottenroth (in this volume) calls Gestaltschlieung, so that we conceive of a COMPLETE ENCLOSURE of what in
reality is partially bounded, as we do with a bowl, if used in the PP the
fruit in the bowl, when some of the fruit is ranging over its edge.
Another case of idealizing of the boundary conditions may occur by
bounding of a MASS, as with SCALE entities referred to in the idioms within
the reach of, at the height of, where the SCALE noun functions as a classifier
of some background, by which a certain amount of the per se unbounded
MASS is particularized. The same holds for MENTAL or SEMIOTIC CONTAINER
entities, which per se are U N B O U N D E D STATES or ACTIVITIES in the temporal
domain. Here volitional action or time is presumed as delimiting background for individuation, as in in the conviction/hope that, in the style of
the 1950s.4
With raisins in the dough, a nail in the wood, a red thread in the fabric,
the delimiting background is the presumed function of the per se unbounded MASS; this function only operates on an individuated part.
Moreover, against this background knowledge the individuation of the
369
may be easily achieved, by mentally referring to a lexical substitution, such as raisins in the cake, a nail in the board, a red thread in the
cloth; in analogy to the COUNT nouns cake, board, cloth reference to an
individual part of dough, wood, fabric occurs.
In its extreme form, idealization can be compared to spatial relations
defined within the mathematical field of topology (cf. Talmy 1983: 261 f.).
What the ideal meaning of the preposition requires is satisfied by drifting,
adapting, projecting or completing the gestalt of the trajector in dependence of the landmark. For instance the shape and magnitide of the
landmark may be irrelevant. What counts is the geometric relation.
However, we have seen that idealization can only be achieved within
certain limits. An example was the preposition along. It is not only its
trajector which must fulfil certain standards, but also its landmark. The
latter may not go under a certain treshold of extension as in ?the items
along the left margin. So along seems to be inherently restricted with
respect to magnitude on behalf of its etymological derivation; originally
it denoted a long extension (cf. The Shorter Oxford English Dictionary).
MASS
(2)
ABSTRACTION
Abstraction is complementary to the process of idealization and consists

in focussing on what is essential in the speaker's view of the spatial scene
whilst ignoring the non-essential properties. Thus the idealized relation
expressed by the PP the trees along the river can only be interpreted or
linguistically encoded, if one abstracts from all details constituting the
real spatial extension of the trees, which would prohibit the idealization
to a line.
The strategies of schematization are driven by several related cognitive
principles which by means of specializing, highlighting, excluding or
prefering certain parts and aspects of a scene instantiate those spatial
properties which are essential in the specific situation (cf. Herskovits
1986: 73 ff.):
The salience principle results in the metonymic relations holding between the parts of an entity or its projections which become functionally
salient in alternative spatial configurations. For instance in the conceptualization of the sentence the children are riding on the bus it is not the
whole bus, which is THREE-DIMENSIONAL, that is salient, but only its TWODIMENSIONAL floor. This becomes selected by the prototypical functional
relation assumed between the landmark entity and the action of riding,
which implies the basic location of the trajector with respect to the
370
landmark: in this case the relation of COINCIDENCE holds; the trajector is

located within the space occupied by the landmark. In this schematization
of an entity's salient shape only some of its spatial properties are focussed
on, while the rest is neglected, as its function is irrelevant within this
configuration. The latter holds, e.g., for the vertical dimension, which
might well become salient if the function of the bus as a double-decker
were in the foreground. The INTRINSIC ORIENTATION of bus is also irrelevant
in this configuration, which one might well move into the foreground in
a configuration designated by the bus is going towards London. By uttering
the sentence The children are riding on the bus the original native speaker's
intention was to refer to a bus established by a platform, which by its
normal function of carrying passengers rendered its SURFACE to be its
salient shape. This can be explained etymologically: originally busses were
platforms in England, and while the concept may not be vivid anymore,
because busses have become closed objects in England too, the expression
has been kept and relates the action of riding to large vehicles in general,
as in ride on a plane, ride on a train, ride on a ship. Another explanation
is that the concept of a SURFACE is enforced by the verb, the original and
typical meaning of which is that of riding on top of an animal. From
this use of the verb the use of passengers being carried by an artificial
means of transport has been derived. Thus by the salience principle the
cross-linguistic relation of COINCIDENCE, which has been assumed as the
prototypical functional relation, is specialized by the relation of SUPPORT
in English.
The typicality principle implies the designation of a spatial configuration dependent on the typical relations existing between the entities in
normal situations. In our example The children are riding on the bus the
discourse situation which is about transport in traffic makes the hearer
assume that the default relation corresponding to the typical situation
holds, in which the children are located inside of the bus and not on top
of it.
Likewise the typicality principle constrains the possible figure-ground
reversals, and relations as expressed by the following sentence are not
schematized: ?The bus is under the children.
The typicality principle also makes possible the interpretation of an
entity's INTRINSIC orientation as the default case if information to the
contrary is lacking. Some entities possess an intrinsic spatial orientation
in analogy to the normal orientation of the human body within the
gravitational system. For instance human bodies, churches and other
371
buildings have an intrinsic frontal orientation and a prominent vertical

axis, which the human conceptualizer associates with these objects independently of the surrounding objects and the dimensional grid of the
earth, unless the normal position of a human being or church is violated.
In this case contextual orientation may override the intrinsic one.
The tolerance principle controls the permitted degree of flexibility in
the use of prepositions, i.e., it guarantees that the spatial configuration
designated by the preposition is within the range of permitted deviations
from the ideal meaning. This is done by constraining the pragmatic
conditions under which expressions chosen by the speaker are adequate.
The tolerance principle may, for instance, control whether the idealization
of trajector and landmark to a point, as it corresponds to the use of the
preposition at, is adequate with respect to the specific position of the
speaker. The tolerance principle also controls the specific range of PLASTICITY of a relation, PLASTICITY is the general possibility of stretching the
boundaries of a spatial schematization type with respect to the range of
possible scenes conforming to it. Thus, in dependence of the gestalt,
function and magnitude of the related objects in front of may be used
even if the located entity is not exactly in front of but also slightly beside
another entity.
The same holds for the corresponding relation of nearness in the vertical
dimension expressed by the preposition over:
(9)
?A bird is flying over the house.
(10)
A plane is just flying over the house.
The relations are illustrated in Figure 1.

>VN
Figure 1. The plasticity the relation expressed by over
372
The relation of nearness may only be schematized in dependence of the

magnitude of the trajector. Although moving in the same distance from
the house as the plane, the bird is too small it would normally not be
perceivable so that the relation of vertical nearness cannot be stretched
as far as to include the bird's position. The same relation, however, may
be stretched, to include the plane as the trajector in the respective
utterance situation. The significance of the relation as to this configuration can be explained by the plane's magnitude and its perceivable noise.
This sentence might for instance be uttered, when the noise of the plane
disturbs a telephone conversation.
The relevance principle implies that dependent on the communicative
goals the speaker can choose semantically alternate prepositions for one
and the same spatial configuration. The communicative goals depend on
the speaker's viewpoint, which in turn depends on the utterance situation.
The following example may illustrate this principle. Imagine a scene in
which Mary is inside the building of a supermarket. If the speaker were
far away from the scene he would designate Mary's location by saying
Mary is at the supermarket, thus expressing that he idealizes the threedimensional extension of the supermarket to a zero-dimensional point
which is associated with the supermarket as an INSTITUTION: by using the
preposition at he asserts that Mary's position coincides with that of the
supermarket. If the speaker, however, were himself on the premises of
the supermarket, he would rather designate Mary's location by the
sentence Mary is in the supermarket, thus referring to the three-dimensional extension of the building of the supermarket, which in this relation
functions as an ENCLOSURE. This shows that different utterance situations
result in disjoint conceptualizations and hence different expressions of
the same physical scene. The different spatial conceptualizations of the
landmark rest on different implicatures about the rest of the scene which
sometimes are not overtly expressed (cf. Talmy 1983: 230). Depending
on which background the speaker associates with the scene, he schematizes the spatial configuration into different relations, by selecting different aspects to have communicative relevance.
Thus by means of different contextual requirements these principles
drive the process of schematizing spatial scenes: the typical function
assumed for an entity in a specific configuration enforces its salient shape
and at the same time excludes all other shapes. Speaker and hearer make
default assumptions about the normal relations between the parts of a
373
scene. The indexical meaning of the preposition is controlled by the range

of possible proportions between the entities determined by some background scale. The communicative relevance determines which relation is
schematized for a specific spatial configuration.
5.2. Schematization and the lexicon

Although we associate many individual concrete entities with a canonical
representation of the shape which belongs to the whole entity, there are
many parts of our environment with which we do not intuitively associate
a canonical shape representation. Take the example of a meadow: it is
exclusively a matter of the speaker's viewpoint whether a meadow functions as a two-dimensional area the length of the grass and other plants
being irrelevant, or as a three-dimensional layer of earth, possibly with
trees, air and grass. Yet, this semantic distinction is not part of the
denotation of meadow, but is a matter of contextual modulation. This
can be shown by using one and the same token in different meaning
relations as in the following example, which does not turn out to be
semantically anomalous:
(11)
She walked through the meadow on which long grass was growing.
In this sentence the meadow is semantically constrained in two different

ways: the preposition through enforces a three-dimensional view on
meadow, while the modifying relative clause, which is attached by the
preposition on, makes us view the meadow as a two-dimensional area
covered with grass. As both senses can be activated without any zeugmatic
effect, meadow is lexically vague with respect to its dimensional properties.
The same holds for cupboard in the flowers on top of the empty cupboard,
or for river in the ship on the river which I walk along. This confirms the
validity of the principles of schematization: it is the context which specializes an entity's spatial properties: salience, indexicality and relevance
have been shown to be contexutally established.
As to our ultimate goal of N L P system design it seems that lexical
information either guarantees the potential of schematization in being
maximally general and inclusive, that it excludes impossible idealizations
of an entity, or that it provides the default assumption about an entity
or situation. The context then has the function of specializing the generalized lexical information or of relating less typical meanings to the
defaults by regular metonymy rules. For instance, the speaker's distri-
374
bution of attention, which partitions a scene into foregrounded trajector

and backgrounded landmark is lexically constrained by assigning per
default typical functions to an entity, as for instance VEHICLE OF TRANSPORT FOR HUMAN BEINGS as the typical function of bus, which as the
landmark unifies with the verb ride, which is assigned as one of its senses
t h e ACTION BY WHICH HUMAN BEINGS ARE TRANSPORTED IN A VEHICLE. T h e
noun child is inherently H U M A N , which means that this lexical property

may not be contextually defeated (these definitions are given in the Oxford
Advanced Learner's Dictionary; they have been slightly paraphrased).
The assumption of the lexically encoded typical function of the bus also
enforces its salient shape and the verb, by its lexically encoded senses
provides the discourse situation, which in turn determines the relevant
relation of COINCIDENCE, specializing as support in English, to be schematized.
What is also lexically significant is the intrinsic orientation of entities,
which may only be overridden by contextually violating circumstances
which are highly unusual. In terms of types of knowledge and their
processing we assume the following lexical encodings:
the spatial disposition of an entity, such as +

its typical function
its inherent vz. intrinsic properties
constraints on idealization
SHAPE, + DIMENSION
By these properties we can unify those component parts of a given

configuration which are in conformity with each other. Thereby a specific
schematization is achieved by recursive specialization of the spatial dispositions. In the case of defeating conditions, which contrast with the
default assumptions encoded in the lexicon, unification on the basis of
the lexically encoded defaults fails. It is by general pragmatic principles,
which generalize over unusual situations, that overriding of some of the
lexical properties is attempted in dependence of those properties which
have to be kept constant.
5.3. The process of schematization and its translation

The process of schematization results in the asymmetrical relation between trajector and landmark. The relational concept the trajector's
spatial disposition is designated by the preposition.
375
As indicated in the last section the trajector's site, path or orientation

are the result of a complex interaction between the parts of the spatial
scene on which the speaker's attention rests.
In this section we want to model this complex interaction by applying
the strategies and principles which constitute the process of schematization to our example sentence The children are riding on the bus, which
translates into German Die Kinder fahren im Bus. On the basis of this
constitution we want to model the mutual development of the meaning
of the assumed parts of the scene: the parts condition each other by
relating information from different semantic resources which control their
typicality, salience and relevance. We have chosen this example, as the
preposition's meaning becomes relevant in the translation into German.
Whereas in German busses provide CONTAINERS for passengers, in English
a passenger is conceptually located on the SURFACE of the bus.
Our ultimate aim to model the process of translation with this sentence
divides into the following steps:
(1) to profile the language-specific relation of the source language in
analysis;
(2) to abstract from the language-specific relation a relation which we
assume to be constant across languages;
(3) to hand this abstract relation together with the trajector-landmark
configuration and the discourse situation over to the target language and
to transfer at the same time the lexical units which are semantically
constrained by the preposition;
(4) to unify the transferred relation with the language-specific meaning
of the lexical units taking part in the relation and thereby to specialize
the abstract relation into the specific one of the target language;
(5) to generate the target language expression out of the relation obtained
in (4).
The model to be achieved via these subgoals should be the precondition
for its implementation in a translation system.
In Figure 2 we represent how these goals are achieved in the process
of translating one schematization type into another. We start with the
source language sentence as input, the language-specific relation of which
is decoded first according to the principles of schematization, which
become active in the assumed situation. The analysis of the English input
sentence starts with the lexically encoded information:
(1) The lexical unit child is associated with the semantic categories
(lexsemcat) HUMAN and MOBILE; the first category is inherent, which means
376
that, according to cognitive constraints, this information may not be

defeated.
(2) The lexical unit ride denotes the action by which human beings are
transported by something or someone; so human beings are required as
the subject in the selectional restrictions of the verb and the means of
transport as the object.
(3) The preposition on predicates the relation of SUPPORT implying CONTIGUITY.
(4) The noun bus denotes a movable, but immobile entity; it typically
refers to its function as a LARGE VEHICLE for HUMANS; typical properties
establish the default, which applies in typical situations; properties applying in other situations may be derived. For instance another lexical
property of bus is its intrinsic front, from which all localizations determined by it may be derived.
We assume the schematization process to operate on this lexical information in the following way:
(5) The lexical information of the verb TRANSPORT of someone by
someone or something is assumed to provide the discourse situation.
(6) The speaker's distribution of attention is constrained syntactically
and semantically; the children as HUMANS and MOBILE entities are in subject
position and hence interpreted as the trajector to be localized with respect
to the bus, which as IMMOBILE ARTEFACT is in object position and is
interpreted as the landmark.
(7) On the basis of the lexically encoded typical function of the bus and
in dependence of the discourse situation, the shape of a SURFACE which
is salient and the floor as the part which is relevant in this situation are
idealized.
(8) The subject is unified with the verb's selectional restrictions.
(9) The verb is unified with the bus' idealized two-dimensional floor as
SURFACE and the relational meaning of SUPPORT is generated, which unifies
with the ideal meaning of on. SUPPORT is defined as that idealized relation
in which the trajector is contiguous with a LINE or SURFACE of the
landmark and the landmark supports the trajector, formally represented
as:
CONTIGUOUS ( L M , T R ) & SUPPORT (SURFACE ( L M ) ,
TR)
(10) The English concept of SUPPORT is subsumed by the relation of

COINCIDENCE, which is the cross-language relation (C) to be transferred
to the target language. The discourse situation is transferred as part of
377
the sentence meaning. The trajector-landmark configuration is also transferred by default. In Figure 2 the transfer of these "interlingually constant" properties is represented by a broken line.
(11) By lexical transfer all "referential" lexical units are transferred into
the target language by unifying the semantic features: child translates
into Kind, ride into fahren; bus into Bus.
Now the schematization process operates on the language-specific
lexical and compositional rules as well as on the transferred information
in order to generate the correct target language expression:
(12) In contrast to English in German the salient shape of Bus idealized
from its prototypical function and the discourse situation provided by
the verb is CONTAINER and the surroundings of the whole entity are
relevant in this situation.
(13) The transferred COINCIDENCE relation is unified with the CONTAINER
concept of bus to specialize as the INCLUSION relation, which unifies with
the ideal meaning of the lexical unit in.
By outlining the schematization process with the interpretation of the
English source sentence it should have become clear that by idealizing
the salient shape of the landmark entity to pertain to its relevant part,
all other schematizations are excluded from being possible. Another
schematization, though unusual, would, for instance, be that in which
the children are sitting on top of the bus; this relation would not unify
with the relevant part. Yet another schematization would be that in which
the children are conceptualized as being enclosed by the bus; this would
be the relation of inclusion, which again would unify neither with the
salient shape nor with the relevant part of bus in this situation, but rather
with the shape of a taxi, of which all enclosing sides become salient to
establish a CONTAINER, and in is used, as in the customer in the taxi.
6. Spatial metaphor
In German spatial metaphor is most frequently realized with prepositions
(cf. Wunderlich 1985: 76). Thus prepositional metaphor may be taken to
offer a model for metaphorical processes in general. The basic idea is
that physical entities are directly meaningful as the basic building-blocks
of our life and of any conceptualization related to it. The process of
378
380
profiling a relation between trajector and landmark takes place with the
perception of concrete spatial scenes and with the conception of abstract
scenes, which in analogy to the model of concrete space are schematized
in mental space. That is, we assume abstract mental space configurations
in the domain of time, causation and reason, circumstance, manner,
means etc. (see Dirven in this volume) to be semantic extensions of spatial
configurations relying on physical properties. This implies the hypothesis
that the configurations expressed by prepositions always rely on physical
properties either directly or by meaning extension of one or more physical
properties into some abstract domain. Such an extension may result from
metonymy, metaphor or other function as they are given in detail, e. g.,
by Lipka (1988); Allan (1981); Nunberg (1978); Norrick (1981); Dirven
(in this volume). Applications in Computational Linguistics are reported,
e.g., in Wilensky (1990). Among the most frequently occurring meaning
extensions in abstract text types are INSTITUTIONS and SEMIOTIC contents
or INSTRUMENTS: by semantic extension CONCRETE objects such as buildings and manifestitations of SEMIOTIC contents function, among others,
as CONTAINERS in an ABSTRACT sense when configurating as a landmark
in a relation expressed by the preposition in.
Metonymy and metaphor are productive processes operating on the
lexicon. It is central to the theory of Cognitive Grammar that these
processes are treated only as one case and thus in accordance with the
general phenomenon of semantic extension resulting in polysemy. Semantic extension to new literal meanings and to metaphorical meanings
only differs in the degree of semantic distance between old and new sense.
If a certain semantic distance is exceeded, especially if a domain shift
occurs, the semantic extension is considered as metaphorical.
It is the characteristic trait of metonymic and metaphorical extensions
to denote semantic domains which are distinct from the domain of the
respective source sense. The transfer of meaning from one domain to
another is most productive with spatial expressions because of their
expressive force: our most basic experience is spatial, therefore spatial
expressions are ideal for improving our experience and imagination of
less basic information. Prepositions are one example of meaning extension
from concrete into increasingly abstract domains. The productiveness of
this process clearly shows its linguistic motivation and communicative
function: metaphors are created in order to understand one thing in terms
of another. The less-known is understood and communicated in terms of
the well-known. In particular, the creation of metaphors involves a
381
mapping between clearly distinct domains. This mapping is effected by

creating an analogy between the structure of source and target domain.
In analogy to the schematization of concrete entities we schematize
abstract entities in the domain of human behaviour and conventions. We
use the structure of our spatial knowledge about entities in physical
source domains for the conceptualization of entities in abstract target
domains (cf. Lakoff Johnson 1980: 81). The reason is the knowledge
transfer from concrete source domains into abstract target domains in
order to interpret the latter in terms of the former. Lakoff and Johnson
(1980) see the essential motivation for the creation of metaphors in the
fact that we understand abstract things which seem difficult for us to
interpret in terms of concrete things for which we have simple, obvious
and perceivable interpretations. "... we typically conceptualize the nonphysical in terms of the physical that is we conceptualize the less
clearly delineated in terms of the more clearly delineated" (Lakoff
Johnson 1987: 59). Thus we can draw inferences which we could not
draw on a literal basis (cf. Hobbs 1983). This means that metaphorical
processes are constitutive of our understanding of the world and that we
have to interpret them in this respect. Metaphorical extensions are not
totally unpredictable (cf. Lakoff Johnson 1980: 30 ff.). There exist
well-established regular metaphorical mapping relations by which concrete schematization types of the spatial source domain transfer some of
their properties to abstract target domains (cf. Lakoff Johnson 1980:
30). This is defined as "limited inheritance between concepts" by Martin
(1988: 397) rendered by a certain "view relation". "An adequate account
of metaphor must allow that the primary or original meanings of words
remain active in their metaphorical setting" (Davidson 1978: 34). The
relations of analogy, similarity and coincidence and distance, which
constitute the metaphorical mapping relations between different knowledge domains, have to be specified so that the interpretation of new
metaphors becomes possible. In terms of a computationally viable interpretation of new metaphors this would raise the following tasks:
Our knowledge about specific source domains has to be organized in
gestalt structures as proposed by Lakoff Johnson (1980), which represent types of experience about situations.
The adaptation of these "well-understood" structures to the respective
"ill-structured" (Carbonell 1980: 424) target domains into which knowledge transfer has taken place, the deduction of the conventionalized
associations and their resulting use in the target domain, the definition
in terms of knowledge transferring functions.
382
The generation of novel metaphorical senses by letting the knowledge

transferring functions operate on the gestalt structure of the source
domain.
The investigation of the relations holding between source and target
domain also implies the question as to which properties a metaphorical
expression provides that a literal expression lacks (cf. Carbonell 1980:
424).
The degree of regularity pertaining to the patterns of metaphorical
extension is vital for a dynamic interpretation of non-lexicalized metaphors in a computational environment. Dirven's paper in this volume
illustrates that, though different prepositions show up meanings in the
same domains to an astonishingly high degree, there are also semantic
gaps, that is certain prepositions do not fully exhaust the general semantic
pattern. This shows that predictability is possible, though within certain
limits. The conventionalized associations existing between source and
target domain have to be known for this purpose; these are exactly the
properties which become salient in the schematization types.
It is our assumption that a semantic representation which attempts to
guarantee a successful synthesis and to have explanatory value at the
same time, cannot evade the generalization over the semantic structure
of source and target concept. Still, a major question raises the problem
of efficiency as to the adaptation of the human process of knowledge
transfer to the computational process of knowledge transfer.
Following our basic hypothesis about perceptually driven conceptualization it seems reasonable to proceed from an analysis of the concrete
spatial relations of the source domain, both because the perceivable
properties of physical space are better accessible to linguistic analysis (cf.
Lakoff Johnson 1980: 56), and because as directly meaningful properties they can be considered as axioms. The abstract schematization
types of the target domains may then be derived from the source types
in a second step. Let us illustrate and thus corroborate this procedure
with a concrete example.
In the following examples the NPs refer to the same concrete source
domain:
(12)
Die Frau im Mantel, die Frau im roten Kleid

'the woman in a coat, the woman in a red dress'
The interpretations of these expressions result in the following schematization type, which implies the semantic categories by which the prepositional predicate constrains trajector and landmark:
383
mod
cat = pp
predicate
cat = prep
PLACE =
CONTAINS (TR, LM)
trajector
landmark
cat = np
cat = np
ABSTRACTION = CONCRETE
ABSTRACTION = CONCRETE
IDEALIZATION = 3 - D
BOUNDEDNESS = COUNT
IDEALIZATION =
3-D
BOUNDEDNESS = COUNT
SHAPE = {ENCLOSURE = ENVELOPE}
Figure 3. Schematization type of the woman in a coat
The enveloping property of the landmark implies how something is

represented physically. This is typically transferred to semiotic arguments
as in the following examples:
(13)
das Programm in Prolog

'the programme in prologue'
(14)
die Implementierung in ALEP

'the implementation in ALEP'
(15)
Der Brief ist in Kursivschrift geschrieben.

'The letter is written in italics.'
These examples refer to the abstract domain of semiotic conventions.

This domain is related to the concrete source domain of clothing conventions by the salient shape properties of the schematization type. The
following schematization type represents the properties which get salient
in the target domain of semiotic conventions:
mod
cat = pp
predicate
landmark
trajector
cat = prep
cat = np
cat = np
PLACE =
ABSTRACTION =
ABSTRACTION =
CONTAINS (TR, LM)
{ABSTRACT = SEMIOTIC}
{ABSTRACT = SEMIOTIC}
BOUNDEDNESS = COUNT
SHAPE = {ENCLOSURE = ENVELOPE}
Figure 4. Schematization type of the programme in prologue
384
The metaphorical meaning which this target schematization type represents is implied in the prepositional predicate which is used metaphorically. In the concrete spatial domain it can be considered to function as
the vehicle in being instantiated by a landmark whose salient shape is an
ENVELOPING ENCLOSURE. We have seen that its source domain is clearly
distinct from the target domain, although the semantic analogy which is
obligatory for metaphor interpretation also exists. The domain, over
which the metaphorized relation is predicated in this case the semiotic
domain provides the topic (traditionally called the tenor since introduced as such by Richards 1936), which adopts the SHAPE properties. The
ground, which is the relation of analogy on which the agreement between
topic and vehicle rests and which is necessarry for the interpretation, is
rendered by the knowledge transferring functions existing between the
original and transferred properties. The metaphorical transfer consists of
the transaction of the properties which figure out as salient in the vehicle's
schematization types in the source domain to the domain of the topic.
In our example the preposition in with its landmark can be said to
function as vehicle in that it provides its concrete spatial meaning and
thus transfers it to the topic, the SEMIOTIC landmark, which then is
interpreted as the bearer of the SHAPE property which has figured out as
salient in the schematization type of the spatial source domain.
If we consider our own interpetations of these examples, however, a
common trait of the interpretation of established metaphors becomes
evident: often the metaphorical relation is no longer relevant; the interpretation of the abstract term is no longer dependent on the meaning of
its concrete mediator of spatial meaning. 5
Many or from a radical perspective all abstract meanings have
originated as metaphors. With many established metaphors the metaphoricity is no longer vivid, because the metaphor is frozen or even dead.
Their cultural motivation is no longer relevant for their interpretation,
because their meaning is directly associated with them. The computational
quintessence though is this: Whereas today the language user understands
the abstract relational concept without referring to the concrete source
concept (cf. Gibbs 1984), originally analogical reasoning was to help us
to conceptualize more clearly the function of the abstract relational
concept; this reasoning process has to be traced back by the computational linguist in order to construe target schematization types, in terms
of which semantically abstract uses of prepositions may be interpreted.
In machine translation the correct schematization type, implying the
385
relational concept, will then unify with the semantic relation encoded by
the preposition to be used. This clearly is a case in which the computational interpretation is different from the human one.
7. Conclusion
We have proceeded from the claim that Machine Translation needs to be
conceptually constrained in order to interpret and generate different usage
patterns in different languages. The considereration of the conceptual
constraints, which have been suggested by different Cognitive Grammarians, has led us to a model of the semantic lexicon in which the distinction
of lexical vagueness and lexical ambiguity is given special importance.
All polysemes of a preposition are represented in this lexicon as departing
from an ideal meaning. We have shown how the selection, specialization
and modulation of the lexical meaning of an entity comes about by the
degree of relevance, salience and typicality attributed to the parts of a
scene on which the speaker's attention lies in a specific environment of
the language and the utterance situation. The translational context has
given rise to the possiblity and necessity of distinguishing beside languagespecific relations more abstract relations, which have cross-language
validity.
In translating a simple example sentence containing a prepositional
phrase from German to English we have shown the complex interdepencies existing between the cognitive principles during the process of translation. The example of this schematization type has to be corroborated
by testing the range of its generalization. Only having touched on the
use of prepositions in abstract domains, the very beginnings of its formalization are still an untackled problem. Complex as the bus example
given may be, it is far from being representative of how the meaning of
prepositions is constituted in general. While in this example the discourse
situation is provided intrasententially by the action of riding, it will very
often only be revealed in complex textual structures. This opens an area
for future research which will comprise among other things interaction
with a knowledge base and the search for an adequate formalism with
which these complex relations may be expressed.
386
Notes
1. I wish to thank Johann Haller, Wiebke Ramm, Randall Sharp and Jrg Schtz as well
as many other colleagues for having discussed with me earlier versions of this paper.
Thanks are also due to Angelika and Johannes Mller for their patience in drawing the
graphics as well as to David Horton who has given the finishing touches to my English.
2. Yet we know that this is a rather crude idea of the process of translation, as many
concepts with cultural significance in the source language have to be transferred directly
into the target language or adapted to the knowledge of the target language users. In
this paper, however, we are not concerned with this problem.
3. This is a slightly altered example taken from the EC ESPRIT programme.
4. Cf. Zelinsky-Wibbelt (1991 and 1992) as well as Zelinsky-Wibbelt (forthcoming) for a
detailed discussion of the conceptual bounding and unbounding of masses.
5. According to psycholinguistic evidence (cf. Pollio et al. 1984, Gibbs 1984, Verbrugge
1977) the human interpretation of established metaphors does not proceed from the
source meaning in a two phase interpretation but is rather contextually resolved directly.
That is, under the respective contextual conditions the hearer/reader has direct access
to the metaphorical meaning and does need any regress to the source meaning.
References
Allan, Keith
1981
"Interpreting from context". Lingua 53: 151-173.
Annick, Peter James Pustejovsky
1990
"An application of lexical semantics to knowledge acquisition from corpora".
in: Proceedings from the 13th International Conference on Computational
Linguistics, Helsinki 1990. Vol. 2: 7-12.
Barsalou, Lawrence W.
1982
"Context-independent and context-dependent information in concepts".
Memory and Cognition 10 (1).
1989
Dimensional adjectives. Grammatical structure and conceptual interpretation.
Berlin: Springer.
Boguraev, Branimir James Pustejovsky
1990
"Lexical ambiguity and the role of knowledge representation in lexicon
design", in: Proceedings from the 13th International Conference on Computational Linguistics, Helsinki. Vol. 2: 36-41.
Briscoe, Ted Ann Copestake Branimir Boguraev
1990
"Lexical semantics via lexicology", in: Proceedings from the 13 th International
Conference on Computational Linguistics, Helsinki 1990. Vol. 2: 42-47.
Carbonell, Jaimes G.
1980
"Metaphor: an inescapable phenomenon in natural language comprehension". in: Proceedings of the 18th Meeting of Computational Linguistics: 415434.
Cruse, D. Alan
1986
Lexical semantics. Cambridge: Cambridge University Press.
387
Dahlgren, Kathleen
1988
Naive semantics for natural language understanding. Amsterdam: Kluwer Academic Publishers.
Dahlgren, Kathleen Joyce McDowell Edward P. Stabler
1989
"Knowledge representation for common sense reasoning". Computational
Linguistics 15 (3): 149-170.
Davidson, Donald
1978
"What metaphors mean". Critical Inquiry 5: 31^4-7.
Franks, Bradley Nicholas Braisby
1990
Sense generation or how to make the mental lexicon flexible. [Research Paper
EUCCS/RP-39. Centre for Cognitive Science, University of Edinburgh.]
Frege, Gottlob
1892
"ber Sinn und Bedeutung", Zeitschrift fr Philosophie und philosophische
Kritik N F 100: 25-50.
Geiger, Richard A. (ed.)
1994
Reference in Multidisciplinary Perspective: Philosophical Object, Cognitive
Subject, Intersubjective Process. Hildesheim: Georg Olms Verlag.
Gibbs, Raymond
1984
"Literal meaning and psychological theory", Cognitive Science 8: 275-304.
Gunderson, Keith (ed.)
1975
Language, Mind, and Knowledge. Minneapolis: University of Minnesota Press.
Habel, Christopher Michael Herweg Klaus Rehkmper
1989
Raumkonzepte in Verstehensprozessen. Tbingen: Max Niemeyer Verlag.
Hawkins, Bruce W.
1985
The semantics of English spatial prepositions. [Doctoral Dissertation.]
L. A. U. T. Paper 142. (Linguistic Agency of the University of Trier.)
Herskovits, Annette
1986
Language and spatial cognition. Cambridge: Cambridge University Press.
Hobbs, Jerry R.
1983
"Metaphor interpretation as selective inferencing. Cognitive processes in
understanding metaphor." Part 1 and 2. Empirical Studies of the Arts 1 (1
and 2).
Hobbs, Jerry R. William Croft Tedd Davies Douglas Edwards Kenneth Laws
1987
"Commonsense metaphysics and lexical semantics", Computational Linguistics 13: 241-250.
Hllen, Werner Rainer Schulze (eds.)
1988
Understanding the lexicon. Meaning, sense and world knowledge in lexical
semantics. Tbingen: Niemeyer.
Jackendoff, Ray
1987
Consciousness and the computational mind. Cambridge, Mass.: MIT Press.
Lakoff, George
1982
Categories and cognitive models. L. A. U. D. Linguistic Agency of the university of Duisburg. Duisburg.
1987
Chicago, London: University of Chicago Press.
Lakoff, George Mark Johnson
1980
Metaphors we live by. Chicago: Chicago University Press.
388
Langacker, Ronald
1987
Foundations of Cognitive Grammar. Vol. 1. Theoretical Prerequisites. Stanford:
University Press.
Lehrer, Adrianne
1990
"Polysemy, conventionality, and the structure of the lexicon", Cognitive Linguistics 1: 207-246.
Lipka, Leonhard
1988
"A rose is a rose is a rose: On simple and dual categorization in natural
languages", in: Hllen, Werner Rainer Schulze (eds.), 355-366.
Martin, James H.
1988
"Representing regularities in the metaphoric lexicon", in: Proceedings of the
12th International Conference on Computational Linguistics, Budapest 1988:
396-401.
1976
Language and perception. London: Cambridge University Press.
Moens, Marc Jo Calder Ewan Klein Mike Reape Henk Zeevat
1989
"Expressing generalizations in unification based formalisms", in: Proceedings
of the 3rd European Conference of the Association for Computational Linguistics, Manchester 1989: 174-181.
Myers, Terry Bradley-Franks Nicholas R. Braisby
1989
"Partiality and coherence in concept combination", in: Dunbar, George
Bradley Franks Terry Myers (eds.) Papers from the 1989 Edinburgh Round
Table on the Mental Lexicon. 1-19 (Edinburgh Working Papers in Cognitive
Science.) Edinburgh: Centre for Cognitive Science.
Nirenburg, Sergei Victor Raskin Allen Tucker
1985
"Interlingua design in TRANSLATOR", in: Nirenburg, Sergei (ed.) Proceedings of the Conference on Theoretical and Methodological Issues in Machine
Translation, 1985, New York: Hamilton.
Norrick, Neal
1981
Semiotic principles in semantic theory. Amsterdam: John Benjamins.
Nunberg, Geoffrey
1978
The pragmatics of reference. Bloomington: Indiana University Linguistics
Club.
The Oxford Advanced Learner's Dictionary of Current English
1948
(4th edition) Oxford: Oxford University Press.
[1989]
Pick, Herbert Lind Acredolo (eds.)
1983
Spatial orientation: Theory, research and application. New York: Plenum Press.
Pollio, Howard Michael Fabrizi Abigail Sills Michael Smith
1984
"Need metaphoric comprehension take longer than literal comprehension?"
Journal of Psycholinguistic Research 13: 195-214.
Pustejovsky, James
1989
"Current issues in computational lexical semantics", in: Proceedings of the
3rd European Conference of the Association for Computational Linguistics,
Manchester 1989: xvii-xxiii.
Pustejovsky, James (ed.)
1992
Semantics in the lexicon. Amsterdam: Kluwer Academic Publishers.
389
Putnam, Hilary
1975
"The meaning of meaning", in: Gunderson, Keith (ed.)
Richards, Ivor Armstrong
1936
The philosophy of rhetoric. Oxford: Oxford University Press.
Schweizer,Harro (ed.)
1985
Sprache und Raum. Ein Arbeitsbuch fr das Lehren von Forschung. Stuttgart:
Metzlersche Verlagsbuchhandlung.
The Shorter Oxford English Dictionary
1933
(3rd edition) Oxford: Oxford Clarendon Press.
[1974]
Sowa, John F.
1984
Conceptual Structures Information Processing in Mind and Machine. London,
Addison Wesley.
Steiner, Erich Paul Schmidt Cornelia Zelinsky-Wibbelt (eds.)
1988
From syntax to semantics. Insights from machine translation. London: Frances
Pinter.
Talmy, Leonard
1983
"How language structures space", in: Pick, Herbert Lind Acredolo (eds.),
225-282.
Taylor, John
1989
Press.
Vendler, Zeno
1967
Linguistics in philosophy. London: Cornell University Press.
Verbrugge, Robert
1977
"Metaphoric comprehension: Studies in reminding and resembling", Cognitive
Psychology 9: 494-533.
Wilensky, Robert
1990
"Extending the lexicon by exploiting subregularities", in: Proceedings of the
13th International Conference on Computational Linguistics, Helsinki 1990:
Vol. 2: 407-412.
1953
Philosophical investigations. Oxford: Blackwell.
Wunderlich, Dieter
1985
"Raum, Zeit und das Lexikon", in: Schweizer, Harro (ed.), 66-89.
Zelinsky-Wibbelt, Cornelia
1988
"From Cognitive Grammar to the generation of semantic interpretation in
machine translation", in: Steiner, Erich Paul Schmidt Cornelia ZelinskyWibbelt (eds.), 105-132.
1990 a
Machine translation based on Cognitive Linguistics: What lexical semantics
contributes to the semantic unity of a sentence. Eurotra-D Working Paper 16.
1990 b
"The semantic representation of spatial configurations: A conceptual motivation for generation in machine translation", in: Proceedings of the 13th
International Conference on Computational Linguistics, Helsinki 1990. 299303.
1991
"Token reference vs. type reference: implications for machine translation",
Machine Translation, Eurotra Special Issue 2. 183-192.
390
1992
1994
"Exploiting linguistic iconism for article selection in machine translation",
in: Proceedings of the 15th International Conference on Computational Linguistics, Nantes 1992. Vol. 792-797.
"Reference as a universal cognitive process: a contrastive study of article
use", in: Geiger, Richard (ed.). (Now available as E U R O T R A - D Working
Paper 21.)
4. Image understanding and prepositions

1. Introduction
The aim of our work is to elaborate some formal tools in order to
represent the semantic content of Natural Language (henceforth NL)
expressions referring to space, in the framework of French text analysis.
Many NL markers allow us to describe space but, usually, it is possible
to classify them in two main groups: spatial referents (nouns and adjectives), and spatial relations (prepositions, verbs of movement). We give
some examples below:
(1) referents
a. Le haut de la tour est en ruine.

'The top of the tower is in ruins.'
b. L'arbre est haut de 5 metres.
'The tree is 5 meters high'.
(2) relations
a. Le restaurant est situe entre la banque et le bureau de

poste.
'The restaurant is situated between the bank and the
post office.'
b. Le bateau s'eloigne de la cte.
'The boat is going away from the coast.'
It can be observed that whereas referents (sentences (1 a) and (1 b))

identify a portion of an object, relations (sentences (2 a) and 2 b)) situate
a localized object (called trajector) with respect to a reference object
(called landmark). We take as source data discourses containing sentences
like those above and try to represent their semantic content with the
formal tools we build.
The elaboration of these representations follows a methodology based
on two main points. First, the study is based on a detailed and systematic
linguistic analysis which must highlight and classify the different meanings
of each lexeme, in particular (in the field of spatial expressions) the
distinct spatial configurations it refers to. This survey of the different
meanings has to be as detailed and systematic as possible so as to
formalize the semantics of the spatial expressions covering all their pos-
394
sible uses. Since we are interested in the expression of space in NL, we

naturally set aside all the uses of the lexemes studied which are not purely
spatial or not about concrete objects like, for instance, metaphorical uses.
The second point of our methodology concerns the elaborated formalisms
which, beyond the representation of the semantic content, should have
adequate inferential properties. This means that we want to use the
formalisms to make deductions whose results must be similar to the
reasoning made by a human being. We consider that a "complete"
semantics must aim not only at encoding the content of the lexemes
studied, but also at reproducing the human reasoning involving those
lexemes. Moreover, such a comparison between the results of inferences
and reasoning constitutes a real way to validate the semantics elaborated.
A direct consequence of this inferential constraint is the choice of logic
as an appropriate framework to build our formal tools.
The representation system of spatial entities and relations whose main
lines we explain here has been used to give a formal definition to the
prepositions dans 'in' and sur 'on' and to some Internal Localization
Nouns (henceforth ILN) such as haut 'top', bas 'bottom', devant 'front
extremity', dessus 'top extremity', interieur 'inside', bord 'edge', angle
'corner', etc., which are all lexical elements pointing out the different
portions of an object (Borillo, A. 1988). The two prepositions sur 'on'
and dans 'in' and the ILNs have been studied in the following syntactic
structure which corresponds to a static localization process:
Ntraj etre prep [ILN de] Nland
'Ntraj to be prep [ILN of] Nland'
Some examples of sentences constructed according to this structure
are:
(3)
a. Les armoiries sont sur le devant de la maison.

'The coat of arms is on the front (extremity) of the house.'
b. Le livre est dans I'armoire.
'The book is in the cupboard.'
In this paper we look closely into the semantics of sur 'on' and dans
'in' showing that a three-level approach (geometrical, functional, pragmatic) enables us to appropriately formalize their spatial uses.
395
2. An analysis on three levels

Throughout this paper, we try to illustrate our approach to the semantic
analysis of spatial expressions, namely that we advocate distinguishing
between several types of elements which make up their meaning.
There are two main reasons for this. First of all, these elements clearly
belong to different qualitative categories: separate treatment is therefore
more efficient. Second, studying what is going on within each category
and between them forces us to go deeper into the analysis of the semantics
at stake than if we adopt a global approach. As a result, we progress
towards one of the principal goals of our work, that is, obtaining adequacy between the inferences which can be drawn from the representation
system and the deductions acknowledged by common sense.
The observation on which the first reason is based is founded to a
large extent on the linguistic analysis of the spatial prepositions in two
important works: Herskovits (1982) and Vandeloise (1986). Nonetheless,
we should note that their authors do not conclude as we do and stick
mainly to a global approach.
2.1. Definition of the three levels

Contrary to Leech (1969) and to a certain extent Miller JohnsonLaird (1976), Vandeloise (1986) showed that a purely geometrical representation of the semantics of spatial prepositions is not appropriate. For
instance, if sur 'on' were represented only with the relation of contact,
we could not make the difference between the sentences (4 a) and (4 b):
(4)
a. L'affiche est sur le mur.

'The poster is on the wall.'
b. L'armoire est contre le mur.
'The cupboard is against the wall.'
In the same way, if the inclusion of the trajector in the convex hull of
the landmark fully described the preposition dans 'in' (which is described
this way or simply by inclusion in most of the literature), we would not
explain why in the situation depicted by Figure 1, sentence (5) cannot be
asserted (this last example is taken from Herskovits 1982).
(5)
La mouche est dans le verre.

'The fly is in the glass.'
396
In these examples, the geometrical approach does not take into account
the functional components of the semantics of the prepositions sur 'on'
and dans 'in': for the first one, the notion of support is essential, and for
the second, containment is very important. More generally, the functional
aspects of the relations and entities they describe play a major part in
the semantics of spatial markers. These functional aspects may belong to
the "naive physics group" (as defined in Hayes 1985), as do the relations
of support and containment; or they may belong to the "orientation
group", as do the functionality of a cupboard which makes it possible to
speak of the front part of the cupboard; finally, they may belong to the
"structure group", as do the characteristic of being considered as discrete
or continuous and all the aspects concerning the composition of the
entities.
As in any field of NL, pragmatic phenomena influence the semantics
of the spatial markers, as can be seen in the following example.
(6)
a. La voiture est dans le garage.

'The car is in the garage.'
b. Le moteur de la voiture est dans le garage.
'The car's engine is in the garage.'
c. Le moteur de la voiture 'est pas dans le garage.
'The car's engine is not in the garage.'
From sentence (6 a) the semantics of dans 'in' should enable us to infer

sentence (6 b). Indeed, if (6 a) is true, then (6 b) is also true or at least it
cannot be denied: clearly, here (6 c) is false. Nevertheless, knowing (6 a),
we would not utter (6 b) alone because it somehow "implies" that the
engine is separated from the car, lying somewhere in the garage.
397
Several pragmatic principles can be isolated, as for instance the "principle of maximum trajector" involved in the last example: "do not localize
a part if you can localize the whole, unless you have good reasons to do
so" and its corollary "generally, the entity localized is not a part of a
bigger entity being in the same situation". These principles or conventions
are in fact instances of more general ones that rule any kind of discourse
or dialogue, as Grice's principle of cooperativity which is composed of
four maxims (Grice 1975) (the principle of maximum trajector is just a
particular case of the maxim of quantity), and the search for the highest
degree of typicality. Of course, these principles generate implicatures (as
the above corallary), not inferences. Consequently, these will have to be
represented within a non-monotonic logic. We should also stress that
they need more information than is contained in the text being analysed,
that is, they often rely on world knowledge (here we used the knowledge
that a car has an engine) and extralinguistic context (here we have
assumed a "normal" context, not one, for instance, in which the car
engine has just exploded and someone is asking where it occurred).
To sum up, we have identified three types of elements involved in the
"meaning" of the spatial markers we studied. First, there are of course
the geometrical properties of the entities and the geometrical relations,
which cannot be put aside. 1 Second, we have seen that the functional
properties and relations also play an important part. Third, we have seen
that pragmatic principles together with a representation of (some) world
knowledge and (part of) the context can significantly modify what is
obtained from the elements of the first two types only. Consequently, the
representation system we have built to formalize the semantics of the
spatial markers consists of three parts.
2.2. Organization of the levels

Before we stark looking into each one of the three parts which make up
the representation system, we must say a word on how these parts are
organized.
The geometrical level is at the root of this system. It constitutes a
formal structure where we represent the objective, purely spatial data
present in the text analysed. This data is said to be objective not only
because it does not take into account context and discourse conventions,
but also because it is independent from the function and internal structure
of the entities, which is often induced by the particular point of view the
398
Michel A u r n a g u e L a u r e Vieu
cFb
ibl
icFb
Fb
st
L -
Figure 2. Partial description of a c u p b o a r d (topology)
399
Figure 2 (following)
entity or
spatial referent
symbol
amoire
face a
face a'
corps-face a
interieur-corps-face a
bord 1
interieur-bord 1
face b
face b'
corps-face b
interieur-corps-face b
ar
fa
fa'
cfa
icfa
bl
ibl
fb
f'b
cfb
icfb
autour
interieur-autour
aut
iaut
c(-,-)
ar*
Fa*
Fa
(cfa*, fa*)
p(bl*, fa*)
p(cfb*,fb*)
p(bl*, fb*)
p(fa*, ar*)
p(fb*,ar*)
cFa* icFa
ar*
Fa*
Fa
cFa*
icFa
bl*
ibl
Fb*
F'b
cFb*
bl*
ibl
Fb*
comp-ass(cfa, fa)
comp-ass(bl, fa)
comp-ass(cfb, fb)
comp-ass(bl, fb)
comp-ass(fa, ar)
comp-ass(fb, ar)
Fb
cFb* icFb
icFb
aut*
iaut
iaut
aut*
note: the addition of * to an entity X (notation X*) means that we consider the spatial
referent of that entity X (that is, X* = sref(X)).
400
locutor has chosen when naming the entities. For instance, sentences (7 a)
and (7 b) describe the same geometrical relation (contact and relative
position on the vertical axis) between the same concrete "objects", although the two entities they localize are different, for the first may be
new while the second may be old.
(7)
a. La bague est sur la table.

'The ring is on the table.'
b. Le morceau d'or est sur la table.
'The piece of gold is on the table.'
This difference is not taken into account on the geometrical level, but
on the functional level where all the properties of the entities are represented. As a consequence, and to emphasize the difference between these
two levels, we look upon the elements handled on the geometrical level
as different from the entities handled on the functional level: we call them
the "spatial referents" of the entities. Formally, it is just a trick to say
that all the entities describing the same concrete object, and thus the
same "space portion", make up an equivalence class with respect to the
geometrical relations.
The functional level is of course based on the geometrical level, since
the spatial properties are part of the properties of the entities, and many
functional relations imply geometrical ones (e.g., containment implies
inclusion in the interior). It is at this level that we represent the "crude"
semantics of the spatial markers we have studied. Here and on the
geometrical level, all the information, including the inferred information,
is certain; that is, this information is not defeasible, although some
pragmatic principles may act later as a filter and eliminate "what is not
false but cannot be said" when using the system in a generation mode
(for instance, answering a question).
The pragmatic level relies heavily on functional properties as will be
seen in the fifth section. In addition, it needs two different structures to
represent context and world knowledge, but these will not be dealt with
in this paper, as much work remains to be done on them. The specific
knowledge that the pragmatic level embodies is procedural: either it works
as a filter as we just mentioned, or it produces defeasible information. It
is also here that an occurrence of a spatial expression is interpreted as
an instance of a particular case 2 of this expression. These processes are
not completely formalized at the moment, so in Section 5. we will simply
present in an informal way what in the meaning of the studied spatial
markers we feel belongs to this level, and we will try to justify it. Even
401
though we know that the separation between semantics and pragmatics

is highly controversial, we feel we could use the term semantics to
designate the first two levels and the term pragmatics for the third. We
hope the reader is kind enough not to take offence at this, in case s/he
uses these two terms in a different manner.
3. The geometrical level

As it has been underlined, on the geometrical level we handle the spatial
referents of the entities, that is the space portion determined by their
matter at a given time. At this level we deal with the topological notions
of inclusion, contact, boundary, etc., and with concepts related to projective geometry such as straight line, distance, order on a straight line,
etc.
The actual use of prepositions like sur 'on' and dans 'in' which allows
us to situate a trajector with respect to a landmark shows the relational
nature of the structures handled in the language as opposed to the
absolute spaces used in robotics (where entities are localized by means
of coordinates). This property of space in N L has also been emphasized
in various works related to psychology and linguistics like those of G.
Miller - P. Johnson-Laird (1976), L. Talmy (1983), A. Herskovits (1982)
or C. Vandeloise (1986). In addition, researchers like P. Hayes (1976)
who, in the field of naive physics, try to formalize commonsense reasoning, make a similar assumption. Moreover, two properties of these absolute spaces seem to contradict the structures of space in NL. Whereas,
in a coordinate system, the positions of every entity need to be known
exactly, the spatial information expressed in a text is often partial and
imprecise. Another problem arises from the fact that the variable granularity of space in N L (for instance in N L the same entity can be
considered at one time as a point and later as a volume) is not compatible
with the discrete structure characterizing an implementable coordinate
system, where the minimum units are defined a priori. So, knowledge
representation on the geometrical level will be done within a relational
structure and not within a coordinate system. As a consequence, the
spatial referent of the entities will be taken as primitive elements and not,
for instance, as sets of points of an Euclidian space. Space is therefore
built from the text and not assumed beforehand (this is similar to the
construction of time proposed in Kamp 1979).
402
It must be stressed that, contrary to an absolute space in which all

geometrical concepts can be expressed by using classical theories of
mathematics (for instance, Euclidian geometry), relational structures imply formalizing topology and projective geometry in two independent
modules. We already showed in Aurnague (1991) and Vieu (1991) that,
up to now, every attempt to express those two types within only one
qualitative formalism has failed. Consequently, the geometrical level of
the system is made up of two distinct modules, one dealing with topological information and the other with projective geometry.
3.1. Mereology and Topology

In order to fulfil the relational requirement mentioned above, we formalize the topological information using the individual calculus based on
connection proposed by B. L. Clarke (1981 and 1985). In fact, we use an
extended version of this theory because we modified and completed it so
as to take into account some important spatial concepts in language.
Following this work and the mereological 3 tradition derived from Leonard Goodman (1940), we call the spatial referents of the entities
"individuals". This calculus is based on the sole relation of connection
between two individuals (noted c (X, Y)) to which B. L. Clarke does not
give any precise interpretation. However, in order to provide help in
reading the axioms, definitions and theorems, B. L. Clarke suggests (according to A. N. Whitehead) considering variables as spatio-temporal
individuals, which is the case in our use of this theory, and stating that
two individuals are connected if they "share at least one point" (which
has no real meaning here since we do not interpret the spatial referents
as sets of points). This primitive connection is used to define some
mereological operators as well as Boolean and topological ones. It is
characterized by two axioms, the first stating reflexivity and symmetry
and the latter being a kind of "extensionality axiom" based on a second
primitive, the identity relation = s :
VX c (, ) VX VY (c (X, Y) => c (, X))
VX VY (VZ (c (, X) ^ c (, Y)) ^ X = s Y).
As regards mereology, the disconnection (X is disconnected from Y:
de (X, Y)) between two individuals and the configurations of inclusion
(X is included in : (, Y)) and overlapping (X overlaps : (, Y))
403
are introduced. The inclusion, for instance, is defined stating that X is

included in Y if every connected with X is also connected with Y:
(, Y) = d e f VZ (c (, X) => c (, Y))
Then, the notion of a proper part (strict inclusion) X of an individual
Y can be expressed as:
pp (X, Y) = d e f (, ) ~ (, X)
It must be noted that the part-whole relations (also called meronomies)
used in the language (e.g.: cette roue est une partie de ma voiture, 'this
wheel is a part of my car') are much more complex than the mere
geometrical inclusion because they involve many functional factors depending directly on the entity (and not only on its spatial referent).
Consequently, we define this type of relation on the functional level of
the system.
Two individuals X and Y are overlapping each other if they share a
common individual Z:
(, Y) = d e f 3Z ( (, ) (, ))
Bringing something new to classical mereology (Lesniewski 1927-1931
and Leonard Goodman 1940), the distinction between connection and
overlapping allows B. L. Clarke to define the situation in which two
elements are externally connected:
ec (X, Y) = def c (, )
(, )
A portion X of an individual Y is called a tangential part of Y if there

exists an individual externally connected to both X and Y (the notion
of non-tangential part is defined in the same way).
In the Boolean part of the calculus, the operators sum ( + ), product
(prod) and complement ( ) are defined. We assume here and in the
following that the reader is familiar with lattice structures and Boolean
algebra. This part of the theory is called pseudo-Boolean because of the
lack of a null element which follows from a philosophical postulate of S.
Lesniewski (1927-1931) taken up by A. N. Whitehead (1929) and B. L.
Clarke. According to this postulate, all the individuals handled in the
calculus must have a concrete existence, which is not the case of the null
element.
The third, and last, part of this calculus covers the topological aspects
themselves; it consists in defining the interior of an individual, its closure
as well as the properties of being closed and open. For this definition as
404
well as for the Boolean part, . L. Clarke introduces the fusion operator
f between an individual X and a set of individuals stating that X
constitutes the fusion of if, for every Y connected to X, there exists an
element of which is also connected to Y:
X = s f (a) = d e f VY (c (, ) <> 3Z (Ze c (, Z))).
Furthermore, an axiom ensures, for any set a, the existence of an
individual X corresponding to its fusion (A being the empty set):
Va
(a = A) => 3X X = s f (a)).
Then the interior of an individual X can be defined as the fusion of

its non-tangential parts:
i ( X ) = def f ({ : N T P (, X)}).
The closure of X is calculated by fusing all the individuals which are
not connected to the interior of its complement:
c(X) =
deff({Y:^c(Y,i(-(X)))})
Here too it must be underlined that the topological interior of an

individual does not correspond to what we usually call interior in the
language. For example, when people refer to the natural interior of the
glass, they are usually pointing out the portion of space enclosed between
its sides (where one can put water) and not to the very interior of those
sides which contains only glass and actually corresponds to the topological interior. As the functional aspects play a great part in determining
the natural interior of an entity, we introduce this notion on the functional
level of the system.
From the topological operators "interior" and "closure", open and
closed individuals can be characterized in the following way:
op (X) = d e f X = s i ( X )
cl (X) = d e f X = s c (X)
This topological part of the calculus is called pseudo because of the
impossibility of handling boundary elements present in classical topology.
The principal reasons behind this are the lack of a null element and the
distinction between external connection and overlap configurations.
Finally, B. L. Clarke introduces an axiom which states that every
individual has an interior (i.e., it has a non-tangential part) and constrains
the product between two open elements to be itself open:
405
VX [3Z ntp (, ) VY [VZ [(c (, X) => (, X)) A (C (, Y)

=> (, ))] => VZ [c (Z, prod (, )) => (, prod (, Y))]]].
The individual calculus based on connection, whose main lines we have
just set out, is not sufficient as it is to deal with problems related to the
semantics of space in language. Consequently, we have extended this
theory in order to express some fundamental spatial notions in the
language, such as limit, contact, etc.
We noted that one of the main spatial concepts handled in language,
limit, could not be directly taken into account by B. L. Clarke's calculus.
In order to make up for this lack, we introduce the notion of empty
individual and regard a limit as a tangential part having an empty interior.
The principal reason why limits can be considered as individuals comes
from the fact that we work in a finite domain. If we had added a nonatomicity condition, it would not have been possible to define them in
this way since they would have corresponded to the classical notion of
limit in mathematics. Moreover, we can characterize the limits in terms
of empty individuals because the spatial referents of the objects we handle
are closed and not open. An individual is empty if each of its portions
are open:
vide (X) = d e f VY [ (, X) => op (Y)].
Using this property, we build three types of limits (lim 1, lim 2, lim 3)
through which surfaces, lines and points can be differentiated.
A portion X of Y is a limit 1 if it is a tangential part of Y having an
empty interior and if any tangential part of X is also a tangential part
of Y (which means that everything inside the limit is flush with it):
lim 1 (, Y) = d e f vide (i ()) tp (, ) A VZ [tp (, X) =>
tp (, Y)].
The enveloppe X of an individual Y is the maximum limit 1 of Y:
enveloppe (X, Y) = d e f tp (, ) A VZ [lim 1 (, ) tp (, )].
The notion of limit 1 being thus defined, we can now state that a limit 2
is a "boundary" between two individuals which are themselves of the
limit 1 type:
lim 2 (, Y) = d e f 3Z [lim 1 (, Z) A enveloppe (, Z)
A vide (i (X)) A tp (X, Y) A VT [tp (, X) => 3W [lim 1 (W, Z)
A ec (W, T) A ec (W, Y)]]].
406
Similarly we regard a limit 3 as a "boundary" between two individuals

of the limit 2 category. In short, surfaces, lines and points satisfy the
conditions of a limit 1, lines and points those of a limit 2 whereas individuals of the limit 3 category are exclusively points.
On the basis of this information it is possible to determine the nature
of the spatial referents handled in the system (the predicate con "is
connected" is defined elsewhere in the theory):
surface (X, Y) = d e f c o n () lim 1 (, ) A ~i3Z lim2 (, )
ligne (, ) = d e f con () lim 2 (, ) 13 lim 3 (, )
point (, ) ^ d e f con () lim 3 (, ).
These limit concepts play a great part in the formalization of ILNs
like dessus 'top extremity'; bord 'edge'; angle 'corner', etc.
Contact seems to be another fundamental notion for the processing of
space in language. In fact, the notion of contact gives rise to two different
interpretations. The first one called "strong contact" covers cases in which
the two entities in contact are assumed to "share some boundary points"
(not the points we just introduced, which are in fact individuals). This
type of contact corresponds exactly to the situations described by the
predicate of external connection (ec): scont (X, Y) = d e f ec (X, Y). A
second category of contact (also called "weak contact") which appears
very often in the language, includes all the configurations for which the
entities are supposed not to have point in common although they are
touching together. As it has been stated in (Vieu 1991), this weak contact
is only compatible with a non-classical distance in which two points
situated at a null distance from each other are not necessarily identical.
An entity X is considered to be in contact with an entity Y if the closure
of any open element containing X is connected to Y (a* denotes the
universal individual):
cont (X, Y) = d e f - i c (, ) VZ [[- (Z = s a*) op (Z) A
(, )] => c (c (), Y)].
Although we used this intuition to describe some predicates, the individuals handled in B. L. Clarke's calculus are not interpreted as sets of
points, as we argued for in the introduction of this section. On the
contrary, the author indicates in Clarke (1985) a way to introduce points
in its system as sets of individuals. This definition based on the filter
technique consists in characterizing points through the individuals (connected two by two) "containing them". We proved in Vieu (1991) that
the definition in terms of filters worked correctly for interior points, but
407
was not right for boundary ones. Actually, its application to boundary
points entails that when two individuals are externally connected, any
individual is also connected to them (which is a serious restriction on the
nature of the modelled worlds). We therefore proposed an alternative
definition for boundary points which enables us to represent the limit
concept from those boundary points. Whereas, under the atomicity condition, interior points (whose definition is identical to that introduced by
B. L. Clarke) each correspond to an atom (which generates the whole set
of individuals including it), this is not the case for boundary points.
However, the notion of base of a boundary point is defined as the minimal
set of individuals generating the whole set defining the point. The definition of a boundary of an individual X (a set of X's boundary points)
and that of the base of restricted to X associated with various constraints of minimal thickness are then used to formalize the various limits.
The definition of a limit in terms of individuals introduced above and
this definition based on points provide identical results.
As already mentioned, on the geometrical level of our system we do
not only take into account topological data but we also integrate some
important concepts of projective geometry.
3.2. Projective geometry

In this module we deal with very important notions such as straight line,
oriented straight line, position of a point on a straight line, etc. We have
above all to make it clear that an important assumption of our study is
based on the delimitation of the universe of spatial entities we describe
and process (essentially with respect to their shape). So, at least for the
analysis of ILNs, we restrict the research field to solid, undeformable
and connected objects that also have a normal usefulness. Here we deal
with a class of objects whose shape is roughly parallelepipedic, cylindrical
or spherical.
We associate a system of orthogonal axes (x, y, z) with the spatial
referent of every entity having the object status according to its dimensionality and shape. For example, three orthogonal axes are associated
with a parallelepiped or a sphere but, whereas in the case of the parallelepiped these axes are each parallel to a pair of sides, in the case of the
sphere their position is not exactly determined with respect to the whole
object. 5 As every axis is attached to an entity, it will be identified by
means of a pair (X, Axe) specifying the entity and the axis's name. In
408
this notation, brackets play the part of a function associating the spatial
referent of the axis with an entity and an axis's name. We must underline
that, from the same axis, it is possible to define two oriented straight
lines or directions. Because they depend on the axis, oriented straight
lines are noted with the triplet (X, Axe, Dir) in which the variable Dir
denotes a direction name. Directions handled on the geometrical level
are called abstract because they do not call for any concrete orientation
like top, bottom, left, etc. In fact, the assignment of a concrete orientation
to an abstract direction is a completely functional process that we model
at the functional level of the system. The sorted variable Dir introduced
in the notation above in order to designate an abstract direction takes
its values in the set {p, op}, whereas the constants + and point out
the concrete orientations appearing at the functional level. In the same
way, the variable Axe representing an abstract direction is instantiated
by the values {x, y, z} as opposed to 1, 2 and 3 which denote the vertical,
frontal and lateral concrete axes, respectively. So, these two sorted variables do not take their values in the entity set.
In our representation we project the spatial referent of an entity Y
(noted sref (Y)) on the axes generated by an entity X of which Y is a
part, and we specify with an interval Int, the axis portion filled by that
projection. All this information is integrated in the predicate proj (sref
(), (X, Axe), Int). The variable Int represents the interval [Exta, Extb]
of (X, Axe) resulting from the projection of sref (Y) on the axis (X, Axe).
The spatial referent of an entity is projected on the axes of the entities
which immediately contain it 6 and not on every axis system of the entities
of which it is a part.
We already said that, for the same abstract axis (X, Axe), it was
possible to define two abstract directions (X, Axe, Dir) that we distinguish
with the values or op of the variable Dir. From an abstract direction
(X, Axe, Dir) of an axis (X, Axe), we can build the relation " i n f ' on the
set of projection points expressing that a point precedes another in a
direction. A and being two projection points on an axis, they verify
one of the three following configurations (where eg (A, B) means that A
equals B):
(inf (A, B), Dir)
(inf (B, A), Dir)
eg (A, B).
We use these relations to situate the projection points present on an
axis with respect to each other. We specify the relative locations of the
409
projection points on an axis by means of the predicate rel ((X, Axe),

List) in which the variable List denotes a list of relations like the one we
defined. Thanks to these order relations, some operators between intervals
that are very similar to those introduced by. J. Allen (Allen 1984) for
time can be defined (Aurnague 1991). For example we succeed in calculating that an interval [Exta 1, Extb 1] is more advanced in one direction
(noted as plus dir ([Extal, Extbl], [Exta 2, Extb 2], (X, Axe, Dir))) or
included (inclus ([Extal, Extbl], [Exta2, Extb2], (X, Axe, Dir))) in
another interval [Exta 2, Extb 2].
This system of axes and projections is fundamental in stating the
semantic definition of ILNs as haut 'top', bas 'bottom', dessus 'top
extremity', dessous 'bottom extremity', etc. (we will give them on the
functional level).
Another aspect of projective geometry consists in specifying the notions
of straight lines, oriented straight lines, etc., through definitions based
on the primitive relation "A is situated between and C" between points
(see View 1991). An important aim now is to specify the system of axes
associated with the spatial referent of an entity in terms of this straight
line definition. As the determination of these axes relies mostly on the
shape of the objects, we would want to deal with the information linked
to shape in a qualitative way (as we do for the other types of data), which
seems to be very difficult.
The last point we deal with in geometry is the notion of distance that
we introduce by means of the primitive relation "B is closer to A than C
is". As we pointed out it before, the formalization of the intuitive contact
implies that this distance should be a non-classical one for which two
points situated at a null distance from each other can be considered as
distinct (in case of weak contact).
We can conclude the presentation of this level by saying that we
obtained a complete relational naive geometry. In fact, the information
processed at this level is not restricted to the spatial domain since the
individuals handled here are really spatio-temporal referents of the entities. Consequently, we introduce purely temporal relations which are
similar to those we defined up to now for the space-time structure and
we also express the links between these two data types.
Figure 2 provides a partial description of a cupboard which uses most
of the geometrical tools we have shown up to now. The first table of
Figure 2 provides the various elements appearing in the description and
the symbols by which we refer to them. Some elements (noted in standard
characters) are really entities (and more precisely objects), which means
410
that a speaker can individualize them. Other elements are simply spatial
referents whose presence is needed for the topological consistency of the
spatial structure. The second table shows the connections between the
different spatial referents (note that all the information is encoded with
the primitive of connection). These connections allow us to calculate, by
means of the definitions and axioms of the theory, the inclusions and the
external connections between spatial referents. Later, on the functional
level, the system deduces part-whole relations 7 for the entities having the
object status. Lastly, the system determines (by using the limit definitions)
the nature of some spatial referents (surfaces, lines or points).
As we showed in the second section, geometrical properties are insufficient to grasp the semantics of spatial expressions and, consequently,
we need to take into account on a second level the functional characteristics of the entities.
4. The functional level

At this level we deal with the orientation process, plural structures as
well as some concepts of naive physics such as stabilization or containment. As all these properties directly concern the entities themselves, we
handle variables which represent entities and not simple portions of space
(as is the case on the geometrical level). These are designated through
the "function" sref. It is also at this level that we introduce the "crude"
semantic definitions of the spatial expressions we study.
4.1. Orientations
We already mentioned that only abstract directions are handled in the
geometrical module. The orientation process consists in mapping an
abstract orientation on a concrete one. Since the assignment of a concrete
orientation depends on properties of the object itself or of its context, it
must be formalized on the functional level. This phenomenon appears
very clearly in intrinsic orientation configurations for which the functions
of the different parts of an entity give it a canonical position in space
(with respect to the vertical axis, the user, etc.), entailing the intrinsic
orientation. For instance, the intrinsic front of a computer is derived
directly from the usual position of its screen. Therefore, we represent the
411
orientation by means of the predicate orient ((X, Axe, Dir), (A, D), V)
which specifies the coincidence between the abstract direction (X, Axe,
Dir) of X (e. g.: the direction (ar, x, p) of the cupboard in Figure 2) and
the concrete orientation (A, D) (e.g.: (1, + ) which denotes the top or
the upper orientation). 8 The third term of the predicate points out the
entity responsible for this mapping process, which can be the entity X
itself (in this case variables V and X are identical and the orientation is
intrinsic) or another entity in the context (V and X denote distinct entities
and the orientation is deictic). 9 This predicate is used to elaborate various
rules for the calculation and processing of orientations. For example, we
can deduce an intrinsic orientation of an entity X from the fixedness of
one of its parts Y or state that every vertical deictic orientation is induced
by gravity. This last property is expressed with the following rule (in
which the contextual entity g symbolizes gravity):
VX VAxe VDir VV [[orient ((X, Axe, Dir), (1, D), V) (X = V)] =>
orient ((X, Axe, Dir), (1, D), g)].
Because we restricted the type of entity processed by the system, we
introduce, again for the analysis of ILNs, some conditions on orientations. First, the texts studied are "instantaneous" in the sense that the
entities described as well as the locutor do not change in position with
respect to one another. Moreover, we make the hypothesis that an entity
is oriented at one moment by a unique locutor. The result of the instantaneous constraint (which induces the immobility of both the entity and
the orientor) can be stated by means of the two following postulates:
VX VAxe 1 VDir 1 VAxe 2 VDir 2 VV VA VD - [orient (X, Axe 1, Dir 1),
(A, D), V) orient (X, Axe 2, Dir 2), (A, D), V)
[-1 (Axe 1 = Axe 2) - | ( D i r l = Dir 2)]]
VX VAxe VDir VV VA 1 VD 1 VA 2 VD 2 [orient (X, Axe, Dir),
( A l , D l ) , V) orient ((, Axe, Dir), (A2, D2), V)
[ - | ( 1 = 2) (D 1 = D 2)]].
The first constraint states that the same concrete orientation cannot
be assigned to two different abstract directions whereas the second one
prevents the same abstract direction from coinciding with two distinct
concrete orientations. In both cases, those multiple orientations can be
the result of a movement of the entity or of the orientor-speaker.
Another aspect related to orientation which we deal with on this
functional level consists in determining the central interval of an axis. As
it is shown in Vandeloise (1986), such a determination does not only
412
depend on geometrical properties of symmetry because it is greatly conditioned by functional features. After identifying this central interval, we
can define two poles surrounding it, each in one of the two directions
associated with the axis.
All these orientational tools are used in the semantic definitions of
ILNs as haut 'top'; bas 'bottom'; devant 'front extremity'; etc., expounded
in Aurnague (1991) and allow us to model the component of the semantics
of sur 'on' relative to the position of the entities on the vertical axis. For
instance, we represent the semantic content of the ILN haut 'top' with
the following formula:
VX VAxe VDir VV VN [orient ((X, Axe, Dir), (1, + ) , V)
natur_or ((X, Axe, Dir), (1, + ) , V, )] =>
haut (, , ) = d e f ds pole (, (, Axe, Dir))
VZ [ds_pole (, (X, Axe, Dir)) => part (, )].
This definition states that when the abstract direction (X, Axe, Dir) of
an entity X receives a deictic or intrinsic (variable N) upper orientation,
the maximum entity Y situated in the pole whose abstract direction is
(X, Axe, Dir) constitutes the top of X. In this definition, it is clear that
functional aspects such as orientation directly arise in the semantics of
spatial expressions. This observation confirms the assertion of E. Lang
who, in his work about dimensional adjectives (Lang 1990) in which the
same functional aspects of orientation appear, writes that "the semantic
analysis of these linguistic items has to be justified in terms of nonlinguistic conceptual structure formation."
4.2. The structure of the entities

On the geometrical level, there is only one relation to link a part to a
whole, namely the spatial inclusion (noted p). As a result, it is impossible
to describe a spatial referent as having a specific internal structure.
However, we often give a particular structure to the entities, sometimes
just by choosing the noun phrase used to refer to an object. For instance,
the rice is conceived as a continuous entity whereas the rice grains is
definitely a discrete one (i. e., a collection). At the same time, there exist
several part-whole relations (also called meronomies) between the entities:
"member-collection" (e. g., Texas is a member of the USA), "componentassembly" (e. g., this wheel is a part of this car), etc. This diversity alone
precludes their direct identification with spatial inclusion. Besides, mer-
413
onomies cannot be represented on the geometrical level, because they

rely on functional aspects of the entities. In particular, the relation
"member-collection" cannot be reduced to a mere inclusion as, among
other things, it is not transitive (The USA are a member of U N O whereas
Texas is not).
We felt that dealing with the internal structures of the entities was part
of our work for several reasons. Firstly, ILNs designate parts of a whole,
i.e., they describe a kind of meronomy; in addition, we will see in the
next section that the preposition dans 'in' can be used to describe meronomies. Secondly, structural relations help explain what the links are
between the different entities used to refer to the same physical object.
We wanted to be able to differentiate these entities; at the same time, it
is desirable to express how they are related. At last, we can mention that
on the geometrical level, any sum of individuals is a new individual
whereas few of these sums correspond to actual entities. This can be
explained by the fact that we conceptualize as entity only what has some
kind of internal structure; the knowledge of these structures then takes
on another sort of importance.
4.2.1. The plural structure

The notion of collection is important in NL: most 1 0 plural noun phrases
(e.g., Jean et Marie 'John and Mary'; les arbres 'the trees') refer to a
collection, and many singular noun phrases, too (e. g., le couple Dupont
'the couple Dupont'; la foret 'the forest'). Two types of relations are
associated with this notion: the relation "member-collection" (e. g., between one tree and the forest) and the relation "subcollection-collection"
(e. g., between a smaller group of trees and the forest). To represent them,
we have taken up the lattice structure introduced by G. Link in his "Logic
of Plurals and Mass Terms" (Link 1983: 311-316).
In this structure, also called the "plural" lattice, the entities denoted
by singular noun phrases in the discourse analysed are represented as
atoms, whereas the entities denoted by plural noun phrases are nonatomic elements. The latter are put in relation to their members and to
their sub-collections by the ordering relation (noted <;) of the lattice. 11
It must be noted that a collection referred to by a singular noun phrase
appears as an atom in this structure, therefore its members are not directly
related to it. Still, this atomic entity and the corresponding plural entity
(e. g., the forest and the trees) are linked through the existence of another
entity: the portion of matter making both of them up. Portions of matter
414
are atoms in the plural lattice, but form themselves a new semi-lattice 12
structure a "matter" semi-lattice where the new ordering relation
(noted < ) is interpreted as "is a portion o f ' . Obviously, the two ordering
relations are not unrelated: if we have < i y then we have h (x) < h (y),
where h is the homomorphism between semi-lattices that gives each entity
its corresponding portion of matter. This function h, restricted to the set
of entities which are portions of matter (a subset of the set of the plural
lattice's atoms), is the function identity. The portions of matter are real
entities people can refer to by what is called mass terms (e. g., the water
in my glass, the gold of my ring), and the matter semi-lattice, together
with some other tools, enables G. Link to deal with properties specific
to mass terms such as cumulative reference. 13
The fact that singular noun phrases referring to collections are not
directly represented as non-atomic elements is a very useful characteristic
of G. Link's formalism. Indeed, the two entities the trees and the forest
are distinct: the forest, not the trees, implies that the collection of trees
is closely built-up in space. In the same way, there are several possibilities
of grouping elements in a collection. G. Link gives the example of several
decks of cards: the decks and the cards are two plural entities describing
the same portion of matter, and they are different since, in particular,
counting them gives different results. Furthermore, the distinction between plural and atomic collections makes it possible to account for the
non-transitivity of the relation "member-collection": Texas can be a
member of the plural entity grouping all of the American states (Texas
and Alabama and ...), itself linked to the atom the USA by the fact that
they have the same "portion of matter", while the USA is a member of
the plural entity grouping all of the U N O nations and without Texas
being a member of UNO. In addition, when a collection is described by
a singular noun phrase which is a mass term (e. g., the rice in this bowl),
its identification with the corresponding plural entity (e. g., the rice grains
in this bowl) would be disastrous because, as it is claimed by many
philosophers and linguists (Quine 1960 and Bunt 1985), mass terms do
behave as if they were describing fully continuous entities.
However, E. Bach has emphasized the difficulty in (Bach 1986) for any
entity to pick the right entity corresponding to the portion of matter
which makes it up. He gives as an example the entity the snowman which
should be linked by the function h to the snow making up the snowman,
but this portion of matter can also be described as being constituted by
the water making up the snow which makes up the snowman and actually,
there is no reason to stop here since the water itself is made up of
415
hydrogen and oxygen, etc. It must be noted that these portions of matter
are different entities since snow and water have different properties. So,
even though we accept the presence of portions o f matter as atoms of
the plural lattice and the fact that any entity (including other portions
of matter) may be linked to a portion o f matter through a "constitution"
relation, we think that the role o f the matter semi-lattice has to be played
by the spatial structure introduced on the geometrical level.
In our proposition, to be consistent with the pseudo-Boolean structure
of the geometrical level, we have removed the null element from G . Link's
plural lattice and so on the functional level too, we only have a pseudolattice. For this reason, we have chosen to represent this plural structure
in a mereologic way instead of representing it classically according to set
theory. The chosen primitive is < (noted < i in G. Link's logic), the partof relation o f mereology.
The following are the three postulates required:
VX V Y VZ ( ( X < Y a Y < Z ) = > X <
VX V Y ( ( X <
Y < ) X =
Z))
Y)
Va (3X => 3 Y fu' (Y, a ) ) (here too, the Greek alphabet is used
to denote sets of entities);
along with the following definitions:
' ( X , Y ) ^ d e f 3Z ( < <
at' ( ) =
def
V A ( < => =
overlaps
is discrete from
dr' ( , ) = der-' ( , )
)
is atomic
f ( , ) = d e f V Y ( => < )
13 ( < VZ ( => dr' (, ) ) )
is the fusion of the
is the sum of X and Y
=
< })
=
= d e f f (, { , } )
(, { U : U <
~ = d e f f (, { : dr' ( , ) } )
set
is the intersection of X
and ; exists only if o'
(X, Y )
is the complement of
; exists only if X is
not the collection of all
atoms.
It is necessary to link the plural structure of the entities and the spatial
structure:
VX V Y ( X < => (sref ( X ) , sref ( Y ) ) )
416
We mentioned above that an atomic collection and the corresponding

plural collection were linked through the identity of their portions of
matter: in our case, the link would be made through the identity of their
spatial referents. However, this may raise ambiguities because different
plural entities can describe the same concrete object (i. e., can have the
same spatial referent), as seen with the example of the decks of cards.
Moreover, we have seen that in the case of mass terms describing collections, 14 it is not desirable to link them to a plural entity. Therefore, we
introduce the relation "is the collection" that links a singular (i. e., atomic)
entity designating a collection to the right plural entity. This relation is
noted est-coll. It satisfies the following postulates:
VX VY (est-coll (X, Y)
(at' () - i a t ' (Y)))
VX VY (est-coll (X, Y) => sref (X) = , sref (Y))
VX VY VZ ((est-coll (, ) est-coll (, )) => = Z)
4.2.2. Meronomies
With the help of the preceding plural structure, we are now able to define
the various meronomies that can be expressed in NL.
We already mentioned the two meronomies: "member-collection" and
"subcollection-collection". We consider two more, namely "componentassembly" (e. g., my nose is part of my head) and "piece-whole" (e. g., this
shard was part of my cup). Actually, we have shown in Vieu (1991: 163
172) that another pair of meronomies should be added in order to cover
the whole range of part-whole relations: "portion-whole" (e. g., this is a
portion of the strawberry pie mum made) and "substance-whole" (e.g.,
there is butter in this cake), but as the study of mass term properties goes
beyond the scope of this paper, we decided to set them aside.
The relation "component-assembly" has to be distinguished from the
relation "piece-whole" on the grounds that in the first, the part has a
well-defined function with respect to its whole, whereas in the latter, the
part is arbitrarily defined in its whole (e.g., South-West of France, the
top of the ball). In the first case, the function of the part usually gives it
well-defined boundaries, but in the second case the part must often be
designated by describing its boundaries, for instance, using an ILN.
Moreover, a component can be unconnected (for instance, a set of
spanners in a tool box), whereas a piece is only a connected part.
"member-collection":
el-coll (X, Y) = d e f
at' () (( at' () X < )

3Z (est-coll (, ) X < ))
417
"subcollection-collection":
scoll-coll (X, Y) = d e f [iat' () n a t ' ( Y ) X <
= ) ]Z ( at' () est-coll (X, Z)
< ) (~n at' () est-coll (, )
< ) 3 3U (est-coll (, )
est-coll (, U) < U)
"component-assembly":
comp-ass (, ) = d e f at' () est-coll (, )
fonct (, ) (sref (), sref ())
"piece-whole":
mor-tt (, ) = d e f
at' () 3 est-coll (, )
- fonct (, ) (sref (), sref ())
con (sref ())
Finally, we group all four cases in one predicate in order to represent

the part-whole relations in an unspecified way:
part (X, Y) = def el-coll (, ) scoll-coll (, ) comp-ass
(, ) mor-tt (X, Y)
We also need two postulates for the relation fonct (X, Y) to express
that X has some kind of function with respect to Y. This relation is not
deeply analysed in this work, but it is enough for our immediate purposes
to state the transitivity of "fonct" and the fact that if a part has no
function with respect to a whole, it also has no function with respect to
an all-inclusive entity:
VX VY VZ ((fonct (, ) fonct (, Z)) => fonct (X, Z))
VX VY VZ ((part (, ) - fonct (, ) part (, Z)) =
fonct (X, Z))
It is worth noting that we do not have as a theorem VX VY (pp (sref
(X), sref (Y)) => part (X, Y)), in particular, since if Y is a collection, only
its members and its subcollections can be considered as parts.
4.2.3. Classification of entities
We will see, in particular for the preposition dans 'in', that the semantic
study of spatial markers shows the necessity of distinguishing several
classes of entities. These are: the objects (e.g., a tree, a forest, Mary, a
piece of gold), the locations (e.g., a country, a crossroads, the Atlantic
418
Ocean, the Pyrenees, my garden, a forest) and the space portions (e.g.,
the interior of a box, a hole in the sheet, a cave in the mountain, the middle
of (the interior of) the room). 1 5
There are two important facts to note. First, the space portions are
always described with the help of another entity which is an object or a
location. It has been underlined in Section 3. that the spatial referents of
the entities are determined by their physical matter. This is not directly
the case for space portions, for they are immaterial; still, they are functionally determined from some material entity(ies), so their spatial referents are geometrically determined f r o m its (their) spatial referent(s). 16
Second, the same noun phrase can designate different entities. F o r
instance, the forest can be an object: the collection of trees (or even the
collection of all kinds of beings living in that collection of trees) and it
can be a location: the portion of ground on which the trees grow. The
same holds for many n o u n phrases: a city is either a collection of houses,
or a location; a sea or a river is either a water body, or a location; a
mountain is either a heap of earth, or a location.
But not every entity can belong to a definite class. There is no rule to
prevent someone f r o m forming either the plural n o u n phrase John and
the USA in N L or the corresponding plural entity in the plural structure.
However, we m a k e the assumption that every atomic entity always belongs to one of the three classes. This assumption is rather strong because
it implies that there is no singular noun phrase describing a "mixed"
collection such as John and the USA. This may be a drawback of our
formalism, but we did not find any counter-example to this assumption.
As a consequence, we add the postulates:
VX
VX
VX
VX
(at' (X) => (obj ()

(obj (X) => ( - i l o c ()
(loc (X) => ( obj ()
(sp-port (X) => ( obj
loc () sp-port (X)))

~~ sp-port (X)))
" sp-port (X)))
(X) A - i l o c (X)))
Finally, we can state that in a part-whole relation the part and the
whole are of the same type, restricting this postulate to the cases where
the whole is not a plural entity (which could be a mixed entity):
VX VY ((part (, ) at' (Y)) => ((obj (X) A obj ())
(loc (X) A loc (Y)) A (sp-port (X) A sp-port (Y))))
The role of this postulate is in particular to exclude the interpretation
of a mere spatial inclusion between an object and a space portion as a
part-whole relation. In fact, in la boite est dans l'interieur de l'armoire
419
'the box is in the cupboard's interior', the box is certainly not conceived
as a part of the interior.
In addition, it is necessary to modify the definition of the "piece-whole"
meronomy, which was only conceived for material entities (objects and
locations). For space portions, we actually need to take into account the
material entity that determines it. The interior of a drawer of a cupboard
is a part (piece-whole) of the interior of the cupboard: the two space
portions are determined by two objects linked by a meronomy (here,
component-assembly). This is not the case for the interior of a box
situated in the cupboard: the two space portions have no conceptual link,
neither do the two objects. To allow for this modification, we add the
following postulates, where the relation det (X, Y) should be read as "X
determines Y":
VX (sp-port () => 3Y ((obj () loc ()) det (, X)))
VX VY VT VU ((mor-tt (, ) sp-port () det (, )
det (U, )) => (T = U part (T, U))
4.3.
The semantic definition of the prepositions:

some naive physics concepts
4.3.1. The preposition sur

As mentioned in Section 4.1., the position of the entities on the vertical
axis is a very important distinction criterion between the different cases
of sur 'on' according to the kind of configuration they refer to. If the
trajector is placed higher than the landmark (as in sentence (8 a)), this
situation is referred to as sur 1 ' o n l ' . The case in which the trajector is
on the same level as the landmark (as in (8 b)) is called sur 2 'on 2'. Sur 3
'on 3' applies when the trajector is placed lower than the landmark (as
in (8 c)).
(8)
a. sur 1 'on 1' Le livre est sur la table.

'The book is on the table.'
b. sur 2 'on 2' L'affiche est sur le mur.
'The poster is on the wall.'
c. sur 3 'on 3' La mouche est sur le plafond.
'The fly is on the ceiling.'
On the geometrical level, these various configurations give rise to three

types of contact between individuals (cont 1, cont 2 and cont 3 respec-
420
tively). For instance, cont 1 corresponds to the situation where a piece

1 of X is in contact with a piece 2 of , 1 being higher than 2.
This set of conditions is expressed by the following rule in which the
predicate zonecont ( 1, X, Y) denotes the contact area 1 between two
elements X and Y: 17
cont 1 (X, Y) = d e f cont (, ) 3Z 13Z2 [zonecont ( 1, , )
zonecont (Z2, , ) plus haut ( Z I , 2)].
Thus comparing the relative positions of the two contact areas between
the spatial referents of the entities (and not directly the relative position
of those spatial referents), all sur 'on' configurations will correspond to
one of the three cases mentioned above.
Apart from the geometrical characteristics of the relative position of
the contact area, two important functional aspects arise in the semantics
of sur 'on', namely the notions of "comparable categories" and "stabilization". Two entities X and Y belong to comparable categories if they
have similar dimensions; this is noted as catcomp (X, Y). This property
is calculated by comparing the projections of the spatial referents of X
and Y on each axis of the triaxial system associated with an including
entity (we create here a partial order relation on the projections). The
projection on a particular axis can be more relevant than on the others,
depending on the configuration of the entities involved. For example, in
the case of a sur 1 'on 1' (sentence (9 a)) the respective sizes of the trajector
and the landmark according to the vertical axis are relatively little constrained, whereas for usages of the type sur 3 'on 3' it is more difficult to
admit an extension of the trajector along this dimension (sentence (9 b)).
Thus, we use three different "catcomp" predicates (catcomp 1, catcomp 2,
catcomp 3) corresponding to the three configurations of sur 'on'. Furthermore, functional properties and nature of the entities must be taken
into account in the definition of "comparable categories".
(9)
a. Le vase est sur la nappe.

'The vase in on the table-cloth.'
b. *Le lustre est sur le plafond.
*'The hanging light is on the ceiling.'
Stabilization (sometimes referred to as support) is another functional

concept which actually plays a major part in the semantics of the preposition sur 'on'. In our system, the predicate stabilise (X, Y) indicates
that an entity X stabilizes an entity Y and the following postulate shows
that, contrary to the contact relation, stabilization is transitive:
421
VX VY VZ [stabilise (, ) stabilise (, Z)] => stabilise (X, Z).

Any entity stable in essence (e.g.: ground) is called intrinsic stabilizer
and every entity which does not belong to that category has to be
stabilized by another element in contact with it:
VX ~ stabilisateur intrinseque (X) => 3
( = )
stabilise (, ) cont (sref (Y), sref (X))].
Specific postulates are also introduced in order to account for the
interaction between the part-whole relation and the stabilization process.
For example, if a piece of an entity Y stabilizes an entity X, then Y
stabilizes X:
VX VY VZ [part (, ) ^ p a r t (, ) stabilise (, X)] =>
stabilise (, X).
The concept of total stabilization is defined by saying that an entity Y
totally stabilizes another entity X if, obviously, Y stabilizes X and if,
furthermore, every entity disjoint from Y and directly stabilizing X is
itself totally stabilized by Y. These various conditions are expressed by
the following definition:
s t a b j o t (, ) def [stabilise (, ) VZ [[cont (sref (Z), sref (X))
stabilise (, ) ~ i o (sref (Z), sref (Y)] => s t a b j o t (, Z)]].
All these geometrical and functional tools enable us to state the following definition of the sur I configuration:
sur 1 (, ) d e f c a t c o m p 1 (, ) c o n t i (sref (X), sref ())
stabilise (, ).
This definition shows that if X and Y have comparable categories, if
the contact area of X (with Y) is higher than the contact area of Y (with
X), and furthermore if Y stabilizes X, we can then deduce that X est sur
'X is on Y \ This is the case of a person sitting on a chair or a glass
standing on a table.
The configuration sur 2 'on 2' where the trajector is on the same level
as the landmark, is represented by a similar definition, the only differences
to be found in the kind of contact (cont 2), in comparable categories
(catcomp 2) and in stabilization, which, in this case must be total:
sur 2 (, )
def
catcomp 2 (, ) cont 2 (sref (X), sref ())

s t a b j o t (, X).
422
It must be reminded here that a total stabilization of X by Y implies

that every entity disjoint from Y and directly stabilizing X is itself
stabilized by Y. This way, a TV set on a shelf which is fixed on a wall
can be said to be on the wall. On the other hand, if the shelf has legs and
is resting (at least partially) on the floor, the wall does not act as a total
stabilizer (of the TV) and the TV can only be described as being on the
shelf (the shelf being against the wall).
The definition associated with the sur3 'on 3' configuration differs
from that associated with sur2 'on 2' in the kind of contact between the
entities involved and in the predicate catcomp3 (which is the most
restrictive of the various "catcomp" predicates, particularly along the
vertical axis):
sur 3 (X, Y) = DEF catcomp 3 (, ) cont3 (sref (X), sref ())
stab tot (, ).
4.3.2. The preposition dans

The preposition dans 'in', used in the given syntactic structure and in a
concrete spatial sense, can refer to three different configurations between
trajector and landmark.
In the first configuration, probably the main case, dans links a trajector
which is completely included in the interior of the landmark, as in the
following examples. These cases are called "total dans".
(10)
a. L'eau est dans la verre.

'The water is in the glass.'
b. La ville est dans le brouillard.
'The city is in the fog.'
c. Paul est dans I'ile.18
'Paul is in the island.'
d. II y a un trou dans ce morceau de fromage.
'There is a hole in this piece of cheese.'
e. L'ile est dans la mer.
'The island is in the sea.'
In the second one, the trajector is only partially included in the interior
of the landmark. We refer to these cases as "partial dans".
(11)
a. La rose est dans le vase.

'The rose is in the vase.'
423
b. L'arbre est dans une petite cour.

'The tree is in a small yard.'
Finally, dans can be used to describe cases of meronomy; the trajector
is then a part of the landmark. These cases are called "dans part-of".
(12)
a. on cerveau est dans ma tete.

'My brain is in my head.'
b. La Bretagne est en France.
'Brittany is in France.'
c. Paul est dans le jury.
'Paul is in the jury.'
a. "total dans", "partial dans" and the "interior" function

The first two configurations involve the notion of interior in addition to
geometrical relations (inclusion and overlap).
The interior of an entity must be distinguished: the entity and its
interior must not be considered as a whole, since we want to differentiate
between "total dans" and "dans part-of'. It is worth noting that the
interior, being a space portion, behaves in a very different way from the
object: an object included in the interior is not a part of this interior
since it can be removed without altering the interior's identity and shape,
whereas a (connected) object included in another is a part of it, fully
contributing to its nature. However, the interior of an entity is completely
determined by this entity, and we always perceive one with the other.
This fact has led many authors (including us in an earlier work Borillo
Vieu 1989) to consider the interior as being a part of the entity and
then to formalize dans or in as the mere inclusion or overlap between the
spatial referents of the trajector and the landmark.
The interior of an object can be of two types, depending on its ability
to contain another object.
If the object is a "container" (solid, offering resistance to gravity and
lateral movements), 19 then its interior is defined by functional properties,
in addition to geometrical ones. Otherwise, the interior can only be
defined geometrically. The sentences (10 a) and (11 a) are examples of the
"container" case. As Herskovits and Vandeloise have shown, the "place"
of the interior of the glass (the space taken by the interior) cannot be
defined only by removing the "place" of the glass in the convex hull of
the "place" of the glass, which is a geometrical function. Indeed, we have
already seen in Section 2. the situation of a fly and a glass that illustrates
it (Figure 1). Restricting the scope of the convex hull to the containing
424
parts of the object (as Vandeloise does) is not satisfactory either because
in Figure 1, the fly would remain in the glass. We need to take each of
the object's concavities into account and either "close" it or not, depending on its containing capability.
In the second case, the object is not a "container"; this occurs with
collections: la foule 'the crowd', with non-solid objects: I'eau 'water'; le
brouillard 'fog' as in (10 b), and with solid objects occupying a volume
evenly without filling it: I'arbre 'the tree' (as in I'oiseau est dans I'arbre
'the bird is in the tree'). In these cases, the spatial referent of the interior
is determined by an "outline" function. It should be noted that here too,
this function is not simply the convex hull. To begin with, the outline
clings closer to the sides of the object than the convex hull does. What
is more, a sizeable "hole" may be excluded; for instance, if a crowd forms
a circle, what is situated dans le cercle 'in the circle' is not situated dans
la foule 'in the crowd', i. e. the interior of this circle is not a part of the
interior of the crowd. 20 There are also a few cases where the landmark
is not a container, yet it is solid and it is not a collection. For instance,
in la boite est dans le papier-cadeau 'the box is in the wrapping paper',
the paper is not a container but is actually supported by the box. In these
cases, the landmark completely surrounds its interior: the spatial referent
of the interior is a connected part of the complement of the spatial
referent of the landmark.
The interior of an object is used to locate other objects as in sentences
(10 a), (10 b) and (11a). It is also used to locate space portions as in
(10 d). The interpretation of such a sentence is that the space portion
(here the hole) is a part of the interior of the object (here the piece of
cheese), the meronomy being a "piece-whole" case. It is not just an
inclusion, for the hole is "attached" to the object just as the interior is,
i. e., both are determined by the object.
The interior of a location, as expressed in sentences (10 c) and (11 b),
is determined by the location's boundaries and the vertical axis: it is the
space portion situated above the ground portion considered. But its exact
thickness is not determined. We cannot go too high: a plane flying over
a town is not situated dans la ville 'in the town'. We cannot stay too low:
a bird can be flying dans la plaine 'in the plain'. When the same noun
phrase describes both an object and a location (a town, a forest) the
interior of the location includes the interior of the object.
It should be clear by now that the "interior" function does not embody
a unique concept. However, we will not represent it in such detail because
our perspective is not to generate discourses from perception and world
425
knowledge, e. g., to decide what is the interior of a given object. Rather,

we need to understand how the function interior behaves deductively.
For this purpose, examining in detail what is conceived as an interior
was helpful. We were led to the conclusion that for a first step, considering
the interior as a whole was possible, as its deductive behaviour is almost
homogeneous. We must admit that we are leaving some cases aside,21
but we hope that in the future we can model the various notions of
interior more precisely.
In addition to such cases, there are occurrences of "total dans" or
"partial dans" that do not call for the use of an interior. The localization
of an object in a space portion (e.g., la chouette est dans le creux de
l'arbre 'the owl is in the tree's hollow') is just explained by the inclusion
("total dans") or overlap ("partial dans") of the object's spatial referent
with the space portion's spatial referent. The localization of a location
in another location, without any part-whole relation between the two is
slightly more complicated. Sentence (10 e) is an example of such localization; Berlin-ouest etait en RDA 'West-Berlin was in D D R ' is another
one. In these (few) cases, the landmark totally and tightly surrounds the
trajector. Another type of location's interior for situating locations could
be defined, but its interest would be limited because not every location
situated in such an interior would be dans the first location. For instance,
a field in the island is not in the sea. A location is "total dans" ("partial
dans" is not possible) another location if all the boundaries of the trajector
are also boundaries of the landmark.
An important fact to note is that even though it is impossible to deal
with the function "interior" on the geometrical level (containment and
classification of entities are involved in its definition), the preposition
dans cannot be reduced to the relation of containment. There are many
examples without any containment between the landmark and the trajector: in la mouche est dans la vollere 'the fly is in the aviary', the cage does
not prevent the fly from getting out of it, in any direction whatsoever;
in (10 b), the city is unable to move anyway, but the fog does not even
support it.
The definition obtained for "total dans" is:
dst (X, Y) = d e f ((obj () (obj () loc ()) (i (sref (X)),
sref (int (y)))) (sp-port () obj ()
(mor-tt (X, int ()) X = int ())) (obj ()
sp-port () (i (sref ()), sref ())) (loc ()
loc () front (, sref (), sref ())
front (, sref (), c (sref (ground)-sref ()))))
426
The definition for "partial dans" is:

dsp (X, Y) = d e f ((obj () (obj () loc ()) (i (sref (X)),
sref (int (y)))) (obj () sp-port ()
(i (sref (X)), sref (Y)))
These definitions are accompanied by the following postulates describing the function "interior":
VX VY (Y = int (X) => ((obj () loc ()) sp-port ()
det (, ) VZ (Z = int (X) => = ) de (sref (X), sref (Y))
(obj (X) => (i (sref (Y)), preint (sref (X))))))
VX VY VT VU VR ((T = int () U = int () part (, )
rest (, X, R)) => (sref (T), sref (U) + sref (R)))
VX VY VT VU VR ((T = int () U = int () (i (sref (X)),
sref (U)) => (sref (T), sref (U) + sref (Y)))
With the geometrical function "preint" designating the convex hull of
an individual minus this individual, and the relation rest (, X, R)
expressing that R is the rest of the whole, Y, once its part, X, has been
taken out.
b. "dans p a r t - o f '
The preposition dans cannot be used to describe any meronomy. Vandeloise showed that we cannot say *mon nez est dans ma tete 'my nose
is in my head'. We think that this fact is better accounted for by a
"contrast principle" than by the explanation he gives (the trajector and
the landmark should not share any boundary).
This principle expresses that when someone uses a spatial expression
to describe a meronomy, he (as well as the hearer) conceives the parts as
detached from the whole and situates this part with respect to the rest of
its whole. In the case of sentence (12 a), we can see that detaching the
brain from the head creates an interior where we situate the brain.
Detaching the nose from the head does not create such an interior: the
nose cannot be in the head. With this principle, we can even explain why
in sentences like l'escargot/la noix est dans sa coquille 'the snail/the nut
is in its shell', the whole is described as being in one of its parts! Note
that only "total dans" is then possible between the part and the rest of
the whole. The same principle can be found in many sentences corresponding to the syntactic structure Ntraj est sur ILN de Nland, such as
la poignee est sur le devant de l'armoire 'the handle is on the front of the
427
cupboard'. The relation etre sur (here, "sur2") really links the handle
and the front of the cupboard without its handle.
However, this principle is only required for a "dans part-of" describing
the meronomies "component-assembly" and "piece-whole", and only
between objects. In all other cases every occurrence of a meronomy can
be described by the preposition dans (as in sentences (12 b) and (12 c)).
The definition of "dans part-of' is then:
dpt (X, Y) = d e f part (, ) ((obj () (comp-ass (, )
mor-tt (, ))) => 3Z (rest (, , ) dst (X, Z)))
4.4. Validation: Adequate inferences

In the introduction, we presented our methodology: we wished to obtain
a semantics allowing the same deductions as those accepted by human
beings. The validation phase, i. e., the control of the adequacy between
the two kinds of inferences, can now be done. We verify here the inferences
a definition leads to by itself as well as some of the inferences obtained
by combining the definitions of several markers or cases of the same
marker.
4.4.1. Transitivity of the meronomies
We have distinguished several meronomies on the grounds that we felt
they were describing different situations. It is important to verify that
these meronomies are really different in our system, i. e. that they behave
in a different way. The part-whole relation has classically been taken as
a unique relation and its transitivity assumed. But this has been proved
wrong in Winston Chaffin Hermann (1987).
The relation "member-collection" is not transitive in our system, and
it is consistent with its behaviour in NL as we have already seen:
Le Texas fait partie des USA. and Les USA font partie de l'ONU. =f>
'Texas is part of the USA.'
'The USA are part of UNO.'
*Le Texas fait partie de l'ONU.
*'Texas is part of UNO.'
This relation can be combined with the relation "subcollection-collection":
VX VY VZ ((el-coll (, ) scoll-coll (, Z)) => el-coll (X, Z))
428
This is verified in NL. For instance, we have:

La Belgique fait partie du Benelux, and Le Benelux fait partie de la CEE. =>
'Belgium is part of Benelux.'
'Benelux is part of the EEC."
La Belgique fait partie de la CEE.
'Belgium is part of the ECC."
The relation "subcollection-collection" is itself transitive, as are the
relations "component-assembly" and "piece-whole". We can even combine the relations "piece-whole" and "component-assembly" (in this order
only), the result being another relation "piece-whole". But not every
mixed combination is valid. As shown by Winston Chaffin Hermann, we cannot mix "component-assembly" and "member-collection".
This is also the case in the system:
Le bras de Jean fait partie de Jean.and Jean fait partie du jury. =t>
'John's arm is part of John.'
'John is a member of the jury.'
*Le bras de Jean fait partie du jury.
""John's arm is part of the jury.'
Note that in this case, we do not even have a "component-assembly"
or "piece-whole" relation between John's arm and the jury, since the jury
is a collective entity.
The study of the various combinations and their validity can be found
in Vieu (1991: 293-296).
4.4.2. The ILNs and sur
The definitions given for several ILNs and for the preposition sur 'on'
under the constraints of "instantaneity" and "unique locutor", result in
the two following inferences:
VX VY VZ VN ((dessus (, , ) sur 1 (, )) =>
3U (haut (U, , ) sur 1 (X, U)))
For instance, as common sense requires:
La boite est sur le dessus de l'armoire. =>
'The box is on the top extremity of the cupboard'.
La boite est sur le haut de l'armoire.
'The box is on the top of the cupboard.'
VX VY VZ VU VN ((devant (, , ) sur 2 (, )
derriere (U, , N)) => ~ i s u r 2 (X, U))
429
For instance:
L^tiquette est sur le devant de la boite. =>
'The label is on the front (extremity) of the box.'
L'etiquette n'est pas sur le derriere de la boite.
'The label is not on the back (extremity) of the box.'
The demonstrations of the theorems, along with several others of the
same type are rather complex, in particular due to the comparison of
contact areas in the semantics of sur. They are fully described in Aurnague
(1991: 247-259).
4.4.3. The transitivities of dans

First of all, studying the transitivities of dans shows that it was necessary
to distinguish three cases: "total dans" is (in some cases) transitive,
"partial dans" is not and the transitivity of "dans part-of" depends on
the transitivity of the underlying meronomies. For instance:
"total dans" is transitive (between two objects and a location):
Paul est dans la maison.
and La maison est dans I'ile. =>
'Paul is in the house.'
'The house is in the island.'
Paul est dans I'ile.
"partial dans" is not transitive; moreover, the combination of a
"total dans" and a "partial dans" is not valid:
L'abeille est dans la rose.
and La rose est dans le vase.
'The bee is in the rose.'
'The rose is in the vase.'
*L'abeille est dans le vase.
*'The bee is in the vase.'
Apart from the distinction in three cases, we have taken into account
the class of each argument in the definitions of dans. This was important,
as the transitivity of "total dans" depends on its arguments belonging to
one or another class.
The combination of a "total dans" between an object and a location
with a "total dans" between two locations is not valid:
Paul est dans I'ile.
and L'ile est dans la mer. =t>
'The island is in the sea.'
*Paul est dans la mer.
*'Paul is in the sea.'
430
The combination of a "total dans' between a space portion and an

object with a "total dans" between two objects is not valid, due to the
presence of a meronomy ("piece-whole") in the first relation:
II y a un trou dans le drap.
and Le drap est dans le tiroir.
'There is a hole in the sheet.'
'The sheet is in the drawer.'
*// y a un trou dans le tiroir.
*'There is a hole in the drawer.'
The complete study of all the possible combinations (about thirty) can
be found in Vieu (1991: 307-313), along with the study of combinations
between dans and meronomies (not only those expressed by dans). As it
has just been seen, there are many combinations which are not valid,
even though many authors have until now taken the transitivity of dans
(or in) for granted, defining it only by the relation of inclusion.
5. The pragmatic level

We have just seen that the adequacy was rather good between what the
system is able to infer, and the deductions a human can make. However,
this adequacy is not perfect, because up to now we have only taken into
account the semantic aspects of the spatial expressions and, as seen in
Section 2., pragmatic phenomena have quite an important influence on
their meaning.
5.1. Restrictions on inferences and additional information

For instance, let us go back to example (6) from Section 2. The system
(made up of the geometrical and functional levels) states that the car's
engine is in the garage, whereas a human would not. Therefore, we have
to introduce in the system the underlying pragmatic principles people
use, in order to filter out the relations inferred "wrongly" (they are not
wrong but cannot be said). At the same time, the same principles can be
used to deduce (by implicatures) more information than there actually is
in the discourse.
We will now explore the pragmatic principles involved in the meaning
of spatial expressions, although we do not claim to be exhaustive. These
principles are not really formalized here, as this part of the work is not
completed at the moment.
431
5.1.1. The "fixation principle"

This principle, first introduced in Vandeloise (1986), expresses that the
typical use of an object "fixes" some of its characteristics. For instance,
the front and the back of a car are "fixed" by the usual direction of its
motion: the front of the car remains the same even when the car is going
backwards. We have seen in Section 4.1. that many intrinsic orientations
are determined in this way.
This principle is also at work when talking about the interior of an
object. The interior of a bottle stays the same even when the bottle is
handled upside down so that it cannot hold any liquid. Moreover, some
of the containing concavities of an object can be rejected as part of its
interior on the grounds that there exists another concavity which corresponds to the "fixed" interior. For instance, even if we accept sentence
(13 a) to describe the situation where an upside-down bottle has liquid
which remains in the hollow of its underside, we will not accept (13 b).
(13)
a. II y de l'eau dans le dessous de la bouteille.

'There is water in the underside of the bottle.'
b. II de l'eau dans la bouteille.
'There is water in the bottle.'
In fact, the inference from (13 a) to (13 b) is valid in the system since
the interior of any part of an object is a part of the object's interior.
Therefore, for the application of dans, we must check whether, in the
world-knowledge base a "fixed" interior is specified for the landmark
and whether the trajector is included in this fixed interior. However,
nothing is certain here: a "marked" context may accept the use of (13 b);
that is why we could not take into account the "fixed" interiors earlier
on the functional level. The same actually occurs for any phenomenon
belonging to the pragmatic level: we can always find a particular context
where the effects of a principle are contradicted.
Here is another (extreme) example of the effects of the fixation principle. The following transitivity, correct on the functional level, must be
ruled out here:
Le crayon est dans la main.
and La main est dans le gant. =>
'The pencil is in the hand.'
'The hand is in the glove.'
?Le crayon est dans le gant.
?'The pencil is in the glove.'
A phenomenon related to this principle must be mentioned. Some
objects are not conceived as containers, so they are not usually used as
432
a landmark for a dans expression. For instance, a table is not a container

although it may have a drawer: the sentence the knife is in the table is
difficult to accept in a non-marked context. Certain restrictions with some
locations are also difficult to explain: one cannot say *dans la plage 'in
the beach', *dans le terrain de football 'in the football field' ...
All of the preceding concerned the restrictive aspect of the fixation
principle. As a matter of fact, it can also generate some additional
information. If sentence (13 b) is present in the analysed discourse, we
will have to assume, in the case of a "normal" context, that the water is
in the fixed interior of the bottle. The same occurs for the sentence the
pencil is in the glove, and indeed for many expressions, we even "choose"
the most suitable part of the interior of an object: Mary is in the car
strongly suggests that she is sitting in the passenger space and not in the
boot. It must be reminded here that this kind of non-certain information
has to be handled by a non-monotonic logic.
5.1.2. The principle of maximum trajector
This principle, a particular case of H. P. Grice's maxim of quantity (Grice
1975), has already been introduced in Section 2. The object localized by
a spatial expression is usually the biggest object we can localize using the
same expression. This principle must be used to restrict the results of the
following inference rule, which is obviously true on the functional level:
VX VY VZ (dst (, ) obj () part (, X)) => dst (, Y))
As we already said, the "principle of maximum trajector" can also be
used to add information, stating that in a normal context, for every
object localized by a spatial expression, any object it is a part of is not
localized the same way.
A word must be said about the application of this principle to spatial
expressions involving a "partial dans". In fact, the contrary should almost
be stated, since the relevance of these expressions decreases as the portion
of the trajector totally included in the landmark gets smaller (in proportion). Several rules governing the acceptability of a "partial dans" must
be taken into account, since, on the functional level, any partial inclusion
between an object and the interior of another is described as a "partial
dans". One of these rules states that if there is a typical function between
two objects, the object that is situated in the other being also a part of
a third one, we cannot express that this third object is "partial dans" the
second. For instance, le doigt est dans le de 'the finger is in the thimble'
433
implies on the functional level, la main est dans le 'the hand is in the
thimble', which is clearly not acceptable and discarded by this rule.
We can then see that the pragmatic level has to apply several rules,
which are contradictory in some situations. For the moment, it is quite
difficult to establish priorities; as said before, the pragmatic level is still
in the process of being studied.
5.1.3. The principle of minimum landmark

As the general purpose of a spatial expression is to help the hearer in
finding the object localized by the expression, the maxim of quantity
requires the use of the best localizing landmark. We could conclude that
the smaller the landmark, the easier the finding (the hearer knowing
where it is). For instance, if Peter has a key in his pocket and Peter is in
the kitchen, it is clear that the key is in Peter's pocket is much better than
the key is in the kitchen.12 This principle would actually suggest that
modelling any kind of transitivity is helpless.
What really occurs here is that the last sentence is misleading because
Peter (then, Peter's pocket) is not more likely to be in the kitchen than
in another room, and someone looking for the key in the kitchen would
not think about looking in Peter's pocket. It must be noticed that, if
instead of Peter's pocket, the key was in a drawer or a dish, or anything
that does not move around by itself, the key is in the kitchen would be
totally acceptable.
Moreover, such a principle is not desirable because there are some
cases where the position of the smallest landmark is determined by the
trajector itself. For instance, the gift is in the wrapping paper is not a
"good" sentence to help locate the gift.
So, this principle is not exactly what is needed here, although we must
admit that, in general, if the system is to answer a question like where is
X?, it should give the smallest known location. We can then see that
Grice's maxims cannot directly be applied. Each of them must be examined for the particular domain of spatial expressions.
5.1.4. Relevance
Relevance is a vague concept, which could also cover the preceding
principles. We want to consider several phenomena governing the acceptability of a spatial relation describing a situation with respect to
434
another relation (up to now, the principles described were about the
choice of the arguments of a given relation, not the relation itself).
Several relations may compete. For instance, in the case of a bowl
turned upside down over an apple, the prepositions dans 'in' and sous
'under' could both describe the situation. However, sous is choosen
because the apple and the bowl are not in the situation typically described
when using dans, one in which the bowl contains the apple. Containment
is not a necessary condition for the use of dans, but it obviously makes
it more relevant. Similarly, one is likely to assume that the sentence la
pomme est dans le bol 'the apple is in the bowl' describes a situation
where the bowl contains the apple. However, containment is not assumed
when the typical situation does not involve containment: la main est dans
le gant does not suggest that the glove controls the hand's movements.
Here again, we need world knowledge about typical situations. It must
be noted that typicality and relevance are both concepts that rely heavily
on the function of the objects: the function of a bowl is to contain,
whereas the function of a glove is to surround a hand in order to protect
it. Therefore, the existence of the functional level is crucial for the
pragmatic one.
5.2. Additional inferences

A spatial expression can be used to describe a situation where in fact its
semantics is not totally verified. For instance, a book is usually said to
be "on the table" even though the book is on another book, thus not in
contact with the table. The loosened definition of " s u r l " , by dropping
the contact, gives rise to a kind of "inference", namely the transitivity of
"surl".
These inferences are not real ones and could not be included on the
functional level. Because the relation between the two books is not
relevant, one can "forget" about it and put the top book directly in
relation to the table. A. Herskovits shows in Herskovits (1982) that, if
instead of two books one on top of the other, it was a lid on a tea-pot,
it would be impossible to "infer" that the lid is on the table. In this case,
being on the tea-pot, the lid fulfils its function with respect to the teapot, so it is relevant for a lid to be on a tea-pot. Consequently, this fact
cannot be "forgotten". The underlying pragmatic principles involved in
this example are the maxims of relevance and of quantity. If a fact is
relevant (in this case the lid is on the pot), to express a less precise fact
435
(in this case the lid is on the table) somehow implies that the precise fact
is not verified in the given situation.
We must mention that it was impossible to deal similarly with "partial
dans", that is considering that loosening an inclusion into an overlap is
a pragmatic phenomenon, because a "partial dans" is not always used to
focus on the part of the trajector which is entirely contained in the
landmark. For instance, in la rose est dans le vase 'the rose is in the vase',
the part of the rose which is out of the vase cannot be "forgotten" for it
is the main part of it!
Even though the separation of pragmatics from semantics is not universally agreed on, we hope to have shown how and why separation is
needed in our system. We are convinced that without getting into such
detail, it would be almost impossible to formalize the meaning of spatial
expressions, i. e. to model the deductions allowed by common sense.
6. Conclusion
Formal semantics of NL expressions referring to space must integrate
the different aspects which underlie their semantics. As shown with the
analysis of dans 'in' and sur 'on', mere geometrical data is insufficient to
express the semantics of such lexemes, and functional or pragmatic aspects
must be also taken into account. These observations have led us to build
a three-level system (geometrical, functional and pragmatic) with which
we can represent the meaning of spatial expressions and make deductions.
This modular construction thus permits us to come closer to the natural
reasoning expressed in the discourse and, from this point of view, constitutes a real cognitive approach. Beyond simple comparison between
deductions made and reasonings of the speaker, we contemplate now
validating our system (in collaboration with psychologists) by verifying
through various experiments the psycholinguistic anchorage of the concepts and primitives introduced (M. Borillo 1991). It must be stressed
that the class of lexemes for which we gave a formal definition (namely
the prepositions dans 'in' and sur 'on' and ten ILNs) seems to be wide
enough to encompass the principal concepts underlying the expression of
space in NL. Consequently, we think that the semantic content of most
spatial expressions can be represented within the framework of this threelevel system, perhaps with modifications or additions. Another interest
of this modular approach lies in the fact that the geometrical module
436
which is specially suited to N L processing is really independent of it

because the individuals handled there have only spatio-temporal properties. The characteristics of this geometrical structure make it very useful
for other fields of Artificial Intelligence than NL. Indeed, a similar
approach is used in Randell Cohn (1989) to model naive physics
processes such as phagocytosis and the behaviour of a "force pump".
Acknowledgements
The authors would like to thank Mario Borillo for his continuing guidance. They are also very grateful to Andree Borillo for her judicious
advice. This work was carried out at IRIT, within the framework of two
Ph. D. theses, under the direction of Mario Borillo.
Notes
1. Here we are in opposition to Vandeloise's position who takes an all-functional point
of view. But looking closer at his proposition, purely geometrical aspects are included
in the description of some of his "family resemblances".
2. By particular cases, we mean the different definitions given for each expression in
Section 4. For instance, "total dans", "partial dans" or " d a n s / p a r t - o f ' for the expression
Ntraj etre dans Nland.
3. Mereology was introduced by S. Lesniewski (1927-1931) and taken up by various
researchers among whom H. Leonard N. G o o d m a n . However, people often directly
refer to Leonard Goodman (1940), probably for reasons of accessibility and readability.
4. It must be noted that it is not possible to define the closure of the universal individual
a*, because the complement of a*, which should be the null element, does not exist in
this theory.
5. Associating a system of three orthogonal axes to a sphere reveals its three-dimensional
nature and means that, interacting with it, a person can define a three-dimensional
system. However, because of the regularity of its shape, the exact position of the system
of axes with respect to this sphere cannot be determined.
6. An entity X immediately contains an entity Y if pp (sref (Y), sref (X)) and there is no
such that pp (sref (Y), sref (Z)) and pp (sref (Z), sref (X)).
7. As we already said, part-whole relations cannot be reduced to the spatial inclusion and
require various functional notions to be taken into account. We will explain further (in
Section 4.2.2.) how we define, on the functional level, various categories of part-whole
relations that allow us, for example, to distinguish functionally and geometrically well
"delimited" components of an entity (comp-ass (X, Y)) from arbitrary pieces of it
(mor-tt (X, Y)). The relation part (X, Y) covers all the cases.

8.
9.
10.
11.
12.
13.
14.
15.
16.
17,
18.
19.
20.
21,
22.
437
In fact, another orientation predicate noted "orient 1" exists in our system. It is intended
to deal with the relations between abstract directions like those occurring between the
systems of axes associated with the different parts of the same object. With this
predicate, we can model the inheritance process of a deictic orientation between an
object and one of its parts.
Consequently, the nature of the orientation denoted by the predicate natur or ((X,
Axe, Dir), (A, D), V, N) can be computed by simply comparing the oriented entity X
and the orientor entity V in the predicate orient.
There are few exceptions: les arenes 'the amphiteatre' refers only to a single object.
In this structure, it is then impossible to make the difference between a singleton and
its unique member. We could not find any evidence of NL's behaving otherwise.
Only the join operator is useful with respect to mass term properties. This semi-lattice
is non-atomic in G. Link's theory: he assumes that there are no minimum portions of
a given substance.
To fully understand this logic, the reader is strongly invited to read Link (1983). As
for mass terms' behaviour, the reader may refer to Pelletier Schubert (1989), a good
synthesis of the abundant literature on this topic.
This is not the case for any mass term. For example, it is impossible to determine what
is an elementary particle of "dirt", "mud", "milk shake". However, the "set approach"
has been taken by a number of authors such as Laycock (1972).
Actually, there are two more classes of entity: portions of matter and substances. They
are needed to represent mass terms and the two meronomies we have set aside.
The space portions could be considered as being of the same type as spatial referents,
that is, as being no entity, since their spatial referent as well as all of their properties
derive from those of objects or locations. This position was not chosen because we can
talk about space portions on the same level as for other entities, and we can even state
some of their properties, e. g., the cave is dark and humid.
The position on the vertical axis of an element X with respect to an element Y is
calculated by comparing the projection intervals of X and Y on an axis (Z, Axe) (of
an entity Z) deictically oriented as a vertical axis.
With nouns referring to types of location like countries or regions, the use of dans is
impossible and en 'in' has to be used (e. g., Paul est en France 'Paul is in France').
Because this fact cannot be explained on semantic grounds, we will deal with these
cases as well.
This property is actually complex and goes deeply into several "clusters" of naive
physics. For example, friction is involved in the situation of a bulb contained by a
socket.
Here again, to be able to distinguish between several entities having the same spatial
referent can be useful. Note that the circle is not linked to the plural entity which the
crowd is linked to, so it is not a collection.
For instance, from la ville est dans I'enceinte 'the city is in the (surrounding) wall' and
ienceinte est dans les arbres 'the wall is in the trees', the sentence la ville est dans les
arbres 'the city is in the trees' does not follow, and yet our model does infer this fact.
The characterisation of an interior we give actually covers quite well all cases except
the case of an "outline" function applied to a landmark having a "hole" (in this
example, it is the case for the trees).
This example was provided by E. Audureau.
438
References
Allen, James
1984
"Towards a general theory of action and time", Artificial Intelligence 23, 2:
123-154.
Aurnague, Michel
1991
Contribution a I etude de la semantique formelle de l'espace et du raisonnement
spatial: la localisation interne en frangais, simantique et structures inferentielles.
[Unpublished Ph. D. dissertation, Toulouse: Universite Paul Sabatier.]
Bach, Emmon
1986
"The algebra of events", Linguistics and Philosophy 9: 5-16.
Buerle, Rainer Urs Egli Arnim von Stechow (eds.)
1979
Semantics from different points of view. Berlin: Springer.
Buerle, Rainer Christof Schwarze Arnim von Stechow (eds.)
1983
Meaning, use and interpretation of language. Berlin: de Gruyter.
Borillo, Andree
1988
"Le lexique de l'espace: les noms et les adjectifs de localisation interne",
Cahiers de Grammaire 13: 1-22, Toulouse: UTM.
Borillo, Mario
1991
"Semantique de l'espace et raisonnement spatial", in: Proceedings of the
Colloque Sciences de la Cognition. Paris: MRT.
Borillo, Mario Laure Vieu
1989
"Elements pour la formalisation du raisonnement spatio-temporel naturel",
in: Proceedings of the 7th Congres RFIA. 695-709, Paris: AFCET.
Bunt, Harry
1985
Mass terms and model-theoretic semantics. Cambridge: Cambridge University
Press.
Clarke, Bowman
1981
"A calculus of individuals based on 'connection'", Notre Dame Journal of
Formal Logic 22, 3: 204-218.
1985
"Individuals and points", Notre Dame Journal of Formal Logic 26, 1: 61-75.
Cole, Peter Jerry Morgan (eds.)
1975
Syntax and Semantics 3: Speech Acts. New York: Academic Press.
Gabbay, Dov Franz Guenthner (eds.)
1989
Handbook of philosophical logic. Vol. 4. Dordrecht: Reidel.
Grice, Paul H.
1975
"Logic and conversation", in: P. Cole J. L. Morgan (eds.), 41-58.
Hayes, Patrick
1985
"The second naive physics manifesto", in: Hobbs, J. R. Moore, R. C.
(eds.), 1-36.
Herskovits, Annette
1982
Space and the prepositions in English: Regularities and irregularities in a
complex domain. [Ph. D. dissertation, Stanford University.]
1985
Published at Cambridge: Cambridge University Press.
Hobbs, Jerry Robert Moore (eds.)
1985
Formal theories of the commonsense world. Norwood, NJ: Ablex.
439
Kamp, Hans
1979
"Events, instants and temporal reference", in: Rainer Buerle Urs Egli
Arnim von Stechow (eds.), 376-417.
Lang, Ewald
1990
"Primary perceptual space and inherent proportion schema: two interacting
categorization grids underlying the conceptualization of spatial objects",
Journal of Semantics 7: 121-141.
Laycock, Henry
1972
"Some questions of ontology", Philosophical review 81: 342.
Leech, Geoffrey
1969
Towards a semantic description of English. London: Longman.
Leonard, Henry Nelson Goodman
1940
"The calculus of individuals and its uses", The Journal of Symbolic Logic 2,
2: 45-55.
Lesniewski, Stanislaw
1927-1931 " podstawach matematyki" [On the foundations of mathematics], Przeglad
Filosoficzny [Philosophical Review]. Vols. 30-34.
Link, Godehard
1983
"The logical analysis of plurals and mass terms: A lattice theoretical approach", in: Rainer Buerle Christoph Schwarze Arnim von Stechow
(eds.), 302-323.
Miller, Georges Phillip Johnson-Laird
1976
Language and perception. Cambridge, MA: Belknap Press of Harvard University Press.
Pelletier, Francis Lenhart Schubert
1989
"Mass expressions", in: D. Gabbay - F. Guenthner (eds.), 327 407.
Pick, Herbert Linda Acredolo (eds.)
1983
Spatial orientation: Theory, research and application. New York: Plenum.
Quine, Willard
1960
Word and object. Cambridge, MA: MIT Press.
Randell, David Anthony Cohn
1989
"Modelling topological and metrical properties in physical processes", in:
Brachman Lesveque Reiter (eds.), Proceedings of the first international
conference on principles of knowledge representation and reasoning, 357-368.
Talmy, Leonard
1983
"How language structures space", in: H. L. Pick L. P. Acredolo (eds.),
225-282.
Vandeloise, Claude
1986
L'espace en franfais: semantique des prepositions spatiales. Paris: Seuil
Travaux en Linguistique.
Vieu, Laure
1991
Semantique des relations spatiales et inferences spatio-temporelles: une contribution l'etude des structures formelles de l'espace en langage naturel. [Unpublished Ph. D. dissertation. Toulouse: Universite Paul Sabatier.]
Whitehead, Alfred
1929
Process and reality. New York: MacMillan.
Winston, Morton Roger Chaffin Douglas Herrmann
1987
"A taxonomy of part-whole relations", Cognitive Science 11: 417444.
Computing the meaning of localization expressions

involving prepositions:
The role of concepts and spatial context
Simone
Pribbenow
1. Introduction: Localization expressions

One open problem in the area of Artificial Intelligence is the formalization
of a commonsense theory of space (cf. Hobbs Moore 1985; Davis
1990). What is needed is a theory of spatial reasoning that describes how
people solve spatial problems like orientation and pathfinding or manipulation of objects in everyday contexts, and how people cope with the
spatial attributes of objects like form and location. This spatial reasoning
is of special interest because it forms the intersection of different tasks
such as vision, the use of cognitive or geographical maps, physical
reasoning, and last but not least natural language processing. A
commonsense theory of space must provide the representation and processing of the basic concepts that are needed for reasoning. Natural
language expressions concerning spatial aspects give us good access to
the concepts which underly the human dealing with space. Thus theories
for the processing of spatial expressions can be used as building blocks
for a commonsense theory of space. Therefore, it is expected that the
spatial concepts and principles developed in this paper can not only be
used for natural language processing but also for other tasks involving
space.
This paper is concerned with one aspect of spatial reasoning, the
location of objects. It presents a computational theory of German natural
language expressions that describe the location of objects by stating the
spatial relations between them. In the following, I call these expressions
localization expressions thus focussing on their function and abstracting
from their different linguistic manifestations. In their simplest form,
localization expressions look like the expression in (1).
(1)
das Pferd auf

der Rennbahn
'the horse' 'on' 'the race track'
LE
REL RO
442
Simone Pribbenow
The first object, the horse, is the entity to be located or LE for short. 1
This object is located with respect to the second object, the reference
object or RO, in example (1) the race track. The spatial relation REL
between the two entities LE and RO corresponds to the preposition on.
The following three examples (2)-(4) show the whole range of localization expressions by giving typical examples for the different classes of
localization.
(2)
Die Konstanzer Universitt liegt am Bodensee.

'The university of Konstanz is located by the Bodensee.'
(3)
Die Protestkundgebung der Studenten fand vor dem Hauptgebude statt.

'The student demonstration took place in front of the main
building.'
(4)
Anschlieend zogen die Studenten zum Rathaus.

'Afterwards, the students marched to the city hall.'
The first example (2) describes the localization of an object here the
university of Konstanz and the second (3) that of a situation, the
demonstration. 2 The sentence (4) provides information about the motion
of the students by localizing the end of their march at the city hall. The
first two cases (2) and (3) describe static localization, because the entities
involved as LE are not moving, or their motion is not relevant in the
current context. A localization expression is called a dynamic localization
if it localizes the source, goal (as in (4)), route or direction of a path. In
most cases the path is induced by a passive or active motion. But it is
worth mentioning that objects of a special elongated form such as streets
and rivers (see (4 b)) or a collection of objects forming a line (see (4 c))
can also induce a path.
(4)
b. Die Strae fhrt zum Stadtzentrum.

'The street leads to the city center.'
(4)
c. Der Zug der Demonstranten reicht von einem Ende des Campus
bis zum anderen.
'The line of demonstrators reached from one end of the campus
to the other.'
In general, a localization expression describes the position of an entity,

which can be an object, a situation or a path. This is normally done with
Computing the meaning of prepositions
443
respect to another object, the reference object. It is also possible but in

everyday contexts only seldom practable to use coordinates as in (5).
(5)
Unser Schiff liegt auf 3849' nrdlicher Breite, 15 17' stlicher

Lnge.
O u r ship is located at 3849' north, 15 17' east.'
Up to now, a preposition is used in all examples to indicate the spatial

relation between LE and RO. Although prepositions are likely to be the
most important way to denote spatial relations, adverbs provide another
way of describing localization (see (6 a) and (6 b)). Deictic expressions,
in German hier, da, dort 'here', 'there', are one category of adverbs. They
can be used to describe static localizations (see (6 a)), or dynamic ones if
the affix hin '-ward' is added.
(6)
a. Hier fand der Protestmarsch statt.

'The demonstration took place here.'
Other alternatives are adverbs which denote directions as in sentence

(6 b), or correspond to prepositions, such as vorne, vor 'in front', 'in front
of.
(6)
b. Die Studenten zogen stadteinwrts.

'The students marched toward the city.'
In localization expressions using adverbs as (6), the reference object is

not given explicitly but must be inferred from the textual or situational
context. On the grammatical level, prepositions and adverbs are
conflated by the classification of adverbs as intransitive prepositions.
Intransitive prepositional phrases are phrases without explicit internal
argument, in the case of local phrases without explicitly mentioned
reference object. The missing internal argument the RO of the localization expression must be inferred from the context. Assuming that
it is in the processed phrases possible to infer the RO, I will not make
any distinction between localizations described by means of prepositions
and those involving adverbs in the remainder of this paper.
In the following sections, I will present the computational interpretation
of localization expressions. The processing is embedded in a text understanding system with the additional ability to answer questions. The task
of the processing is to build up an internal model of the text which is
similar to a "mental model" in the sense of Johnson-Laird (1983). Localization expressions determine that part of the model that deals with
the spatial relations between objects. The internal model is used for
444
Simone Pribbenow
further problem solving, in our case for answering questions. Other spatial
tasks for which localization facts could be useful are pathfinding or the
generation of spatial descriptions.
In this paper, I will concentrate on the representation and the construction of the internal model. The question answering is only described
briefly; for more details see Pribbenow (1990). I assume a two-level
semantics consisting of a linguistic level concerned with language-specific
features and a conceptual level based on a language-independent set of
(spatial) concepts. This paper concentrates on the conceptual processing
of localization expressions which uses a combination of propositional
and picture-like formalisms. I will not discuss semantic aspects in detail;
examinations which focus on that subject or on the relation between
semantic and conceptual representation can be found for example in
Bierwisch (1988), Wunderlich - Herweg (1991), or Klein (1990).
The next section discusses the basic assumptions which leads to the
design of a system adequate for processing spatial aspects and the spatial
concepts underlying the interpretation of localization expressions. The
third section describes the phenomenon of localization and its formalization in terms of localization areas. Each localization expression is
interpreted by an area which consists of a primary area characterized by
spatial concepts and priorizations, which rely on expectations where the
LE can most probably be found. The fourth section presents the conceptual interpretation of localization expressions based on those areas. My
goal is to show how textual and situational context influences the interpretation of localization expressions and how this contextual influence
can be integrated in the computation of the (conceptual) meaning of
localization expressions.
2.
Design and conceptual basis of the system
2.1. Basic assumptions

The design of a system that computes the meaning of localization expressions is guided by the following three assumptions from linguistics
and psychology:
A. a two-level meaning representation consisting of a semantic and a
conceptual level;
445
B. the existence of a set of language-independent spatial concepts on the

conceptual level; and
C. dual (hybrid) processing on the conceptual level.
The first assumption A. is based on the work of Bierwisch (1983) and
Bierwisch Lang (1987) who present a general paradigm for the processing of natural language, not only for spatial expressions. The theory
assumes two levels of meaning representation: a semantic level that
focuses on the language itself and a conceptual level which is based on
language-independent general concepts. On the semantic level, underdetermined abstract representations are constructed which normally involve
variables. A transformation function translates these semantic representations into conceptual meaning representations, e.g., by instantiating
the variables and interpreting the abstract semantic labels by concepts.
This makes it possible in a two-level semantics to establish a unique
representation for each lexical item (on the semantic level) as well as a
fully specified meaning representation that reflects contextual influences
(on the conceptual level). For my work, the theory of Bierwisch Lang
has the advantage of allowing me to concentrate on the conceptual level
and ignore the semantic level. Thus I can work out the processing relevant
for a commonsense theory of space without having to worry much about
language-specific features.
Another advantage of the two-level approach is that it reflects the
assumed modularity of cognitive behavior. According to the thesis of
modularity, the linguistic system (to which the semantic level is assigned)
exists independently of but in interaction with the conceptual system (to
which the conceptual level is assigned). While the linguistic system focuses
only on the features of language, the conceptual system forms the interface
between various subsystems of human cognitive behavior, e.g., the different perceptual systems, the motor system, the linguistic system, and
so on. To be able to function as an interface, the conceptual system must
include for each domain a set of concepts which is used by the relevant
subsystems but which is independent of each specific subsystem. Assumption B. specifies this observation for the spatial domain: there exists
a set of spatial concepts which is accessed by subsystems with spatial
tasks, e. g., by the linguistic system for processing spatial expressions, but
which is independent of each single subsystem as the linguistic one.
Therefore the spatial concepts can be characterized as language-independent even so they guide the processing of natural language expressions
concerning spatial aspects. The most important concepts for localization
are presented in the next section (2.2.).
446
Simone Pribbenow
Concerning the processing of space, the relevant subsystems besides

the linguistic system are the vision and the motor system. All these
subsystems are seen as independent modules which interact with each
other via the conceptual system by using overlapping sets of spatial
concepts. This interaction between the subsystems via common concepts
can explain why we are able to speak about our motions, describe the
relations between objects we see, or carry out commands concerning
spatial tasks.
The last assumption C. is concerned with the representation and processing of spatial knowledge, e. g., the spatial concepts. To allow for
effective and cognitively adequate processing, a representation formalism
suitable for space is needed. By now it is generally agreed that for solving
certain problems, especially in the spatial domain, mental pictures are
used by humans. Following the dual-coding theory (Paivio 1983), there
exist at least two independent formalisms for representation and problem
solving, one propositional and one picture-like. In the terminology of
Artificial Intelligence, systems consisting of more than one formalism are
called "hybrid" (Brachman Gilbert Levesque 1985).
The picture-like formalism used for representing space should have
properties of mental pictures, e.g., it should be sketchy. One possible
realization are depictorial representations (Kosslyn 1980) which could be
used as a cognitively adequate representation of space (Habel 1987). The
relevant advantage of picture-like and therefore also of depictorial formalisms is that they "simulate" space. Picture-like formalisms automatically ensure topological and geometrical properties of space and each
picture is complete with respect to the spatial relations between all objects
in the picture once it is created. The last aspect makes the computation
of non-explicit relations very efficient using pictures, whereas a lot of
inferences are sometimes necessary to do the same in a propositional
system (see Lindsay 1988 for more details).
Figure 1 shows the design of a system for text understanding and
question answering that respects the three assumptions A.-C. This system
consists of linguistic and conceptual processing, realizing the semantic
and the conceptual level. The interface to the user is given by the input/
output modules. The linguistic subsystem includes the syntactic and the
semantic processing and uses mainly linguistic knowledge such as grammar rules or lexical entries and in some cases also world knowledge, e. g.,
terminological information. The conceptual system consists of a propositional and a depictorial subsystem and uses only the world knowledge.
This knowledge consists of terminological knowledge in particular in-
447
eluding knowledge about object categories, different types of situations,

and (spatial) concepts, knowledge about individual entities, and (procedural) knowledge about the processing.
input
output
linguistic processing
(syntactic and semantic)
conceptual processing
prepositional subsystem
world
knowledge
(terminological
individual
procedural)
depictorial subsystem
L

I
II
_
,7"!
perceptual
system
.
JI
I
7
.
II
motor
system
JI
Figure 1. A system for processing localization expressions
The assumption A. of a two-level meaning representation is reflected

in the separation of linguistic and conceptual processing. The world
knowledge includes the language-independent concepts (assumption B.).
If necessary, the conceptual module can serve as an interface to other
subsystems which are illustrated in Figure 1 by dotted blocks and arrows.
According to C., conceptual processing works on different subsystems:
one propositional, e. g., realized by a logical formalism, and one depictorial, e.g., realized by cell matrices. One possible implementation of
such a system is the LEU/2-system, a prototype developed by the IBM
in co-operation with five universities (see Herzog Rollinger 1991 for
a description of the system and its underlying theories). The work presented in this paper is integrated in the LEU/2-system.
448
Simone Pribbenow
2.2. Basic concepts for localization

The concepts concerning localization form only a part of the conceptual
basis for processing space. They interact with other spatial and nonspatial, e. g., functional, concepts. It should be noticed that the following
set of concepts is most likely not the only possible base for localization. 3
But it is one that is sufficient for computing the meaning of localization
expressions. The concepts under consideration form two classes:
general localization principles which control the process of localization
itself and
localization concepts which are used to describe concrete localizations.
General localization principles

The most important principles are those of the different roles of RO and
LE and of the influence of the situational context. As stated in the
introduction, a localization expression involves two entities, the reference
object RO and the entity to be located LE. For the process of localization
the different roles of LE and RO lead to different views on RO and LE
(see Talmy 1983; Herskovits 1986).4 The reference object forms the basis
of the localization and must therefore be able to function as a landmark. 5
A localization process can only be successful if the position of the RO is
known in advance or can easily be detected. For that reason, the reference
object is normally more relevant and perceptually conspicious, e.g.,
bigger, than the LE. Often, special parts of the RO such as one side are
focussed to narrow the space under consideration. This is done using
conceptualizations of the reference object (see the explanation of object
concepts above). In contrast, the LE is typically a less salient, smaller
object without further conceptualization.
The different roles of the two objects reflect that in general a localization expression cannot be inverted by simply exchanging RO and LE,
as example (7) shows.
(7)
a. die Straenlaterne beim Park

'the lamp near the park'
b. ?der Park bei der Straenlaterne
?'the park near the lamp'
449
In a situation as described in (7), the lamp can be located relative to

the park. But it does not seem to make sense, even if it is grammatically
correct, to describe the position of the whole park by relating it to a
lamp. This is only possible for parts of the park, e. g., a piece of the wall
surrounding the park.
The second general localization principle concerns the influence of the
situational context on the localization process. According to this principle
a localization expression can only be interpreted or generated by taking
into account the concrete situation. Sometimes the contextual knowledge
is obligatory, for example if adverbs or extrinsic expressions (expressions
needing a "secondary reference object" (cf. Talmy 1983: 245) for their
interpretation) are involved. As stated in the introduction, localization
expressions involving adverbs have no explicit RO; it must be inferred
from the context. For the analysis of extrinsic expressions the relevant
secondary RO, e.g., the spatial origo in deictic cases, must be provided
by the spatial situation under consideration. The following section about
localization concepts provides examples of various secondary ROs extrinsic expressions could refer to. In other cases the situational context
as far as it is known influences the interpretation process. Examples are
objects which restrict the possible location of the LE by functioning as
barriers or competing objects (a definition is given in Section 4.2. C).
Localization concepts
For the description of a concrete localization, e. g., given by a localization
expression, two different classes of concepts are needed: object and
distance concepts. The object concepts allow for different views on a
(perhaps complex) object, which are in the following called object conceptualizations. The distance concepts describe the (qualitative) distance
between two objects.
Some of the object concepts describe requirements on an object concerning its form, dimensionality or plurality. Certain spatial relations like
entlang 'along' or um 'around' assume that the LE can be conceptualized
as a line with a special relation to the RO. The dimensionality of the
reference object restricts the choice of the spatial relation. If the RO is
viewed as a two-dimensional object like a field a motion from one end
of the RO to the other is described with ber 'across', whereas if the RO
is seen as a three-dimensional object like a house durch 'through' must
450
Simone Pribbenow
be used. Although natural objects are always three-dimensional, most

objects allow for different conceptualizations of their dimensionality
according to the context. The concept of plurality is only important for
a small number of relations which need a complex RO, such as zwischen
'between' or inmitten 'amongst'.
More important than the concepts for describing requirements are
those object concepts that provide conceptualizations which focus on
parts of an object. The focussing could be done by using the concept of
relevance of parts or the concept of orientation via reference systems.
The conceptualization of an object with respect to the relevance of parts
is controlled by functional concepts such as containment, different kinds
of support or part-of relations. These functional concepts are involved if
certain prepositions like in 'in', an 'at', 'on' or auf 'on', 'upon' are used,
which in some cases express not only spatial aspects but also functional
ones. In these cases, the RO and the LE must be restricted to those parts
which can provide the relevant function. One well-known example of a
conceptualization following the concept of relevant parts of an object is
given by (8).
(8)
a. das Wasser in der Vase

'the water in the vase'
b. der Sprung in der Vase
'the crack in the vase'
As Figure 2 a shows, the water is located between the sides of the vase,
and the RO is conceptualized only as its hollow space. The crack, however,
must be located in the material parts of the vase (Figure 2 b); it seems to
be more a part of the vase itself than an independent object.
a: the water in the vase

Figure 2. Conceptualizations of the RO
b: the crack in the vase
451
deictic point DP
back
right
L Jl.X.i
right -
ieS
rvvv
front
north
*
' front
left
west
back
a: intrinsic
left
b: deixis
east
south
c: geosystem
Figure 3. Different reference systems
The conceptualization of relevant parts can be also used for the LE.
The localization expression (8 c) describes a situation where normally
only a part of the flowers is located inside the hollow space of the RO
and thus only a partial localization of the LE is provided.
(8)
c. die Blumen in der Vase

'the flowers in the vase'
According to the functionality of containment, the part of the flowers

inside the vase must be long enough to prevent the flowers from falling
out. According to pragmatic expectations, the part inside the vase belongs
to the stems and not to the blossoms. Notice that the localization
expression (8 c) would not turn out to be wrong, if these expectations are
not fulfilled, e. g., the whole flowers including the blossoms felt into the
vase.
Another way to focus on parts of an object is given by the concept of
orientation via reference systems. This concept is only used with objects
serving as the RO of a localization expression. A reference system or
reference frame is projected on the object, so that different parts of the
object belong to different directions of the frame and can be selected
with respect to the applied direction. Figure 3 shows a house which is
oriented by different reference systems.
There are two classes of reference systems, the intrinsic and the extrinsic
ones. 6 In intrinsic cases, the RO itself establishes the frame of reference.
The house in Figure 3 a establishes a reference system with six orientations, which form three pairs of opposite directions. The entrance provides
the front side, and opposite the back. The right and left side are induced
orthogonally. The upper and the lower side are given by the gravitation;
they are omitted in the picture. Reference systems with less than six
orientations are possible, for example, a tower provides only upper and
452
Simone Pribbenow
lower side. There exist also objects, e. g., balls, which are not able to
establish any orientation at all by themselves. The most complete list of
how different classes of objects induce intrinsic systems is presented by
Miller - Johnson-Laird (1976: 403).
To establish extrinsic reference systems, a secondary reference object
is needed. If this secondary RO is the origo of the spatial deixis, e. g., the
location of the speaker or listener, than the frame is called a deictic
reference system. Figure 3 b shows how the orientation of the deictic
point DP is induced on the primary RO in a mirror-like way. The part
of the RO focussing the deictic point becomes the front, its opposite the
back and the part focussing the right or left side of the DP becomes the
right or left part of the primary reference object. The vertical axis with
upper and lower sides remains unchanged. Another kind of orientation
is induced by the geographical system, which forms another class of
extrinsic reference systems. In that case, the two reference objects are not
disjoint, but the primary RO is included in the secondary RO (the Earth)
and inherits its four orientations from it (Figure 3 c). Other ways of
inducing extrinsic orientations can be found in Wunderlich Herweg
(1991).
The object concepts can only be used for each object independent from
the other involved in the localization. To describe the relation between
two objects, distance concepts are used. 7 For the processing of localization
expressions, it turns out to be suitable to choose four basic distances:
inclusion, contact/direct nearness, nearness and remoteness.
The concept of inclusion, referred to by prepositions like in 'in' or
durch 'through', says that the LE is included in the space occupied by or
assigned to the RO. Inclusion is normally related to the functional concept
of containment which is supplied for the LE by the RO or of the partwhole relation between them. To be in a contact/direct nearness relation
to the RO, e.g., as induced by an 'at', 'on' or auf 'on', 'upon', the LE
must be near enough to the RO so that an underlying functional relation
like support or an intentional relation like waiting at a cash point can be
possible. If no functional relation is involved in the situation under
consideration, the two objects must touch each other.
The concept of nearness underlying the majority of prepositions, e. g.,
bei 'near' or vor 'in front o f , does not have any underlying functional
aspects. The LE is near the RO if it is in the vicinity of the RO and the
RO can serve as a landmark for the LE. Sometimes the space of possible
near-locations of the LE is called the "region of interaction" (Miller
Johnson-Laird 1976) of the reference object with respect to the LE. If
453
the LE is outside of this region, the two objects stand in a remoteness

relation to each other as denoted by jenseits 'beyond'.
For a coarser classification, the first three concepts inclusion, contact/
direct nearness and nearness are generalized to the concept of proximity
in contrast to the concept of distality which is provided by remoteness.
The validity of the concepts proximity/distality is not restricted to the
spatial domain. The same pair can be found for example in connection
with events (Herweg 1990).
As the following investigations will show in more detail, each localization can be described by a combination of object and distance concepts.
The LE normally has a neutral (that means no) conceptualization,
whereas the RO is conceptualized according to the object concepts of
relevance of parts, orientation via reference systems or also in a neutral
way. Then the adequate distance concept is applied to the conceptualizations of RO and LE. If necessary the other object concepts (form,
dimensionality and plurality) can be used to describe requirements on
both objects.
3.
A first view on the processing of

localization expressions
3.1. The phenomenon of localization and its formalization

The processing of localization expressions is guided by the features which
are assigned to the phenomenon of localization. The theory described in
this paper is based on two characteristic features of the localization
process:
localization as segmentation and
localization as search.
According to the first characterization, the localization of an entity
involves the segmentation of the space under consideration. Under that
view, a localization corresponds to the delimitation and description of
the part of space that contains the LE. The selected part of space is called
a localization area or, more specifically, primary (localization) area. The
segmentation of the space under consideration can be induced by coordinate systems or by the objects contained in the space. Sentence (5) in
the introduction gives an example for the use of a coordinate system.
454
Simone Pribbenow
R R = (R -
O)
R N = R F U RRS U R
^ RLS
Figure 4. Segmentation of a space
But as stated there, the normal way to localize objects is to relate them
to other objects. Therefore, it is more interesting to see how the delimitation and description is done with respect to an object included in the
space under consideration.
The localization concepts presented in 2.2. are used to segment a space
with respect to an object O. Figure 4 shows that an object divides a space
under consideration into the space occupied by the object itself in
the following also described by place (O) and the surrounding space
R.
If the distance concept of nearness is applied, the space R can be split
up into the nearness-area R N and the remoteness-area R r . If the object
is conceptualized with respect to a reference system, the orientation of
can be inherited to the surrounding nearness-area R N . In Figure 4 an
intrinsic or deictic reference system is used to split R N into four possibly
overlapping parts: R F corresponds to the part of R N related to the
intrinsic or deictic front of the object, R r s to that related to the right
side, R B to the back and R LS to the left side.
The second characterization of the phenomenon of localization as
search emphasizes its procedural aspect. According to that characterization, the selected part of space serves as a "search domain" 8 for the LE.
In order to get a suitable search domain, the selected part of space, the
primary area, is combined with expectations about the probable locations
of the LE. These expectations about locations lead to priorizations.
Positive priorizations are parts of the space that are good candidates for
455
places to find the LE within, e.g., the typical locations of the entity to
be located. Thus these pieces of space are preferred in contrast to the
rest, and are the first to be searched for the object to be located. There
is also the possibility of negative priorizations, which exclude parts of
the space from the parts to be searched.
The result of all available positive and negative priorizations is the
delimitation of a part of the primary area. In the following, such a fully
specified localization area is called a search domain. While the primary
localization area contains all possible locations of the LE, a fully specified
search domain contains only the likely ones. So the last kind of localization area can better serve as a basis for search than the primary selected
space. In situations where no priorizations are available, the primary area
and the search domain turn out to be the same. For all other cases it
should be noted that priorizations are based on expectations and not on
certain knowledge. So it must always be possible to withdraw its results
if an expectation turn out to be wrong after new knowledge is added or
after an unsuccessful search. This is not the case with the primary area,
which should be consistent with the given localization expressions.
The two characteristic features of localization lead to the formalization
of the phenomenon given by definition (9).
(9)
LOK ([relevant_parts](LE), G s )
with a. Gg = G Gpi Gp2 ... Gpn
b. R 2 G,
c. G 2 place ([relevant_parts](LE)]
The definition of localization given in (9) is based on the concept of

the search domain. The predicate LOK relates the LE to the relevant
search domain G s . If functional aspects are involved, only the relevant
parts of the LE with respect to the underlying functional relation between
LE and RO are considered. An example is provided by expression (8 a)
in Section 2.2., which describes the partial localization of flowers in a
vase where the stems provide the relevant parts that are located inside
the RO. As indicated by the squares, this conceptualization of the LE
with respect to relevant parts is optional. If there is no functional relation
involved in the situation, that demands such conceptualization, the whole,
that means neutral conceptualized, LE is considered.
The first condition a. describes that the search domain G s consists of
the primary localization area G modified by the priorizations G Pj . The
area G is the result of the segmentation process. Condition b. ensures
that all areas are really a part of the space R under consideration. The
456
Simone Pribbenow
area G is considered as the set of all possible locations of the LE according

to the known (localization) facts. This implies that the space occupied
by the LE or its relevant parts respectively is included in G. This fact is
expressed by the last condition, where place (LE) or place (relevant_parts
(LE)) refer to the space occupied by the LE or its relevant parts. It should
be noted that no condition requires the LE to be included in the search
domain G s . As explained above, the priorizations represent only default
knowledge which could be withdrawn. We only suppose to find the LE
inside the search domain, but can never be sure of it.
3.2. An overview of the interpretation process

Now the question arises, how these localization areas are computed as
to localization expressions. This section gives an overview of the whole
interpretation process. One part of it, the conceptual processing, will be
specified in the following section.
According to the basic assumption A. given in Section 2.1., the interpretation process consists of two levels, the linguistic and the conceptual
one. The conceptual level involves propositional and depictorial processing as stated in assumption C. Figure 5 shows the steps of the process
by the predicates resulting from each processing step.
The linguistic analysis computes static and dynamic localization facts
from the natural language input. 10 Static localization facts are induced
by expressions describing the location of objects or situations (see examples (2) and (3) in the introduction). These facts are represented by
the predicate Lok sem which indicates the localization of an object or a
situation S with respect to the spatial region SR. A spatial region is an
abstract entity that is defined by the reference object RO and the spatial
relation REL extracted from the input or, if necessary, from the textual
or situational context.
Dynamic localization facts are computed from dynamic localization
expressions which describe the motion of an object (see example (3 a)) or
the course of an elongated object like streets or plural objects (see (3 b)
and (3 c)). They are given by an instantiation of the predicate Move (O,
PR) which describes the motion or the course of along the path-region
PR. Path-regions are internally structured regions consisting of a sequence
of simple spatial regions.
Before the computation of localization areas can start, the semantic
facts must be transformed into a canonical form. The transformation is
C o m p u t i n g the meaning of prepositions
localization
expression
457
natural language input
s y n t a c t i c and s e m a n t i c
analysis
ioksem (0,SR)
Loksem ( S , S R )
Move (0, P R )
t r a n s f o r m a t i o n of
static
linguistic level
t r a n s f o r m a t i o n of
localizations
dynamic
of s i t u a t i o n s
localizations
M o v e (0, P R )
Loksem ( S . S R )
Loksem (0, S R )
LOKsem (Patn, 5 R )
area c o n s t i t u t i n g
process
LOK ( O / P a t h , 6 s )
Depiction
conceptual level
Figure
5.
Interpretation process for localization expressions9
done by abstracting from dynamical and situation-specific features so

that only the static localization of objects must be analyzed by the area
constituting process. This restriction makes it possible to use a static
depictorial formalism with single pictures for the conceptual processing
and to ignore temporal aspects on the propositional level. As the examples
(10) and (11) will show, this is achieved by abstracting from all temporal
aspects in dynamic localization and by abstracting from situations in
static localization involving situations. The abstraction process is carried
out by the conceptual pre-processing which computes a canonical form
for all localization facts.
458
Simone Pribbenow
For static localization facts involving objects, no further work has to

be done. The localization of situations is reduced to the (static) localization of certain objects involved in the considered situation. The first
step is to compute the objects involved in the current situation with the
help of world knowledge about the relevant type of situation (compare
Figure 1 and its explanations). As example (10) shows, which of these
objects are located depends on the given spatial region.
(10)
a. Babette fhrt im Labor Experimente durch.

'Babette is carrying out experiments in the laboratory.'
b. Babette fhrt unter dem Abzug Experimente durch.
'Babette is carrying out experiments under the flue.'
In both sentences, the situation "Babette is carrying out experiments"

is located, but with respect to different regions. The world knowledge
about making experiments contains the information that a concrete
situation of that type normally involves person(s) performing the experiments, the instruments needed and the objects or material examined. In
example (10 a) all these entities can be assumed to be inside the laboratory
and therefore inside the spatial region. The localization of the situation
is transformed to the localization of all involved entities. For (10 b) this
is not possible because a flue is to small to cover whole persons. In that
case, only some of the instruments and the objects examined are located
inside the spatial region, but not the woman making the experiments. So
the localization of objects must be determined according to the interaction
between the situation and the region involved.
Dynamic localization facts are translated into the static localization of
the underlying path. This path is the trajectory of the motion described
or the LE itself if the course of an elongated object was described. Parts
of the path like beginning, end or pieces of the route are located according
to the semantic facts. Often more than one phrase is used to describe the
spatial features of a path as in the example (11).
(11)
Die Frau geht aus dem Haus und steigt in ihr Auto.
'The woman went out of the house and got into her car.'
The result of the linguistic processing of (11) are two dynamic localization facts. The first one is a source-motion which starts inside the house
and ends outside of it. The second is a goal-motion which begins somewhere outside the car and ends inside of it. The conceptual pre-processing
creates in the first step two paths which are later combined into one
single path corresponding to the whole motion of the woman. The
459
beginning of the resulting path is located inside the house and its end
inside the car. 11
The results of the pre-processing are one or more localization facts
which describe the static localization of the object or the part of a
path inside a spatial region SR. Each of these facts is taken as input
for an area constituting process which transforms the abstract region into
a fully specified search domain. This is carried out by combining propositional and depictorial means for the first time. Thus the resulting
localization area is a hybrid object with propositional and depictorial
features. The result of the process of localization is a search domain G s
for an object, or a part of a path together with the depiction of this area.
The problem of partial localization of the LE that might lead to the
conceptualization of relevant_parts(LE) is omitted in the implemented
version of the process and in the remainder of this paper. In the following
the LE is assumed to be conceptualized in a neutral way.
4. Conceptual processing
This section examines in detail the specification of the abstract semantic
representation to get a more detailed conceptual meaning representation.
This is done by converting the spatial region SR used in the semantic
fact into a maximally specified localization area, the search domain G s .
While the region is only determined by the reference object RO and the
spatial relation REL, the search domain takes other parameters into
account, e. g., the entity to be located LE. In the first part of this section
I will explain the computation of a search domain in general, while the
second part focuses on those steps in the computation where distinctions
emerging from contextual influences take place.
4.1. Computation of a search domain

According to the characterization of localization in Section 3.1., the
process of area construction involves the segmentation of the space under
consideration in order to establish a primary area and the priorization
of the primarily separated area. The result is a search domain G s which
is composed of the primary area G and the priorization areas Gpj corresponding to the priorization Pi as described by definition (12).
460
(12)
Simone Pribbenow
Gs = G Gpi Gp2 ... Gpn
The way in which the operation is interpreted depends on the kind

of priorization considered (see 4.2.). The computation of the search
domain uses in addition to RO and REL given by the spatial region
other parameters like the LE, the textual context, e. g., the verb used, or
the spatial context, e. g., other objects in the direct surroundings of the
RO. A complete list of the parameters is presented in Habel Pribbenow
(1988). As expected, the construction of a search domain consists of two
steps:
A. the computation of the primary area G and
B. the specification of G using priorizations.
Step A starts with the definition of the primary localization area which
is computed by propositional means, in our case by using rules. According
to Section 2.2., each localization can be characterized by an LE- and a
RO-conceptualization combined with a distance concept applied to these
conceptualizations. As stated in 3.2., the LE is always conceptualized in
a neutral way. So the rules compute an area definition which consists of
a conceptualization of the RO and a distance concept. If not a neutral
one is used the RO-conceptualization refers to the object concepts of
relevant parts or orientation via reference systems. 12 But this definition
provides only a characterization of an area and not a concrete part of
space corresponding to this area. The concrete area must be built up by
a depictorial process which evaluates the spatial concepts of the area
definition. First, the conceptualization of the RO is evaluated by marking
the appropriate part of the object. 13 The subsequent process interprets
the distance concept with respect to the part marked in the first step.
Although the propositional definition does not carry enough information to select a concrete primary area, it is a necessary step in the
computation process because the definition provides the connection to
the conceptual basis. Thus the regularities between spatial concepts can
be used for the processing of areas. For example, the dichotomy of such
concepts as proximity/distality, inclusion/nearness or opposite orientations like front/back can be inherited to pairs of areas defined by those
concepts. For problem solving tasks like the answering of questions, the
valid regularities can be used in an explicit manner, while on the depictorial level they remain implicit.
The primary area is used as the basis for the second step B, the
construction of the search domain. The processing of the priorizations is
the best example of the interaction between the propositional and the
461
depictorial module. What can serve as a priorization in principle is stated

by propositional means. But only the depictorial module can decide what
is applicable in a concrete case and evaluate the priorization (see the
example of negative priorization in 4.2.). As stated in Section 3.1., the
results of the priorization step are only of default status. So it must be
assured that the results can be withdrawn on both the propositional and
the depictorial level. But despite the uncertain status, the priorizations
are important because they correspond to expectations about the location
of the LE and therefore contain the information which makes a search
effective.
Before I examine the individual steps of the processing in more detail,
I will give a small and rather simple example to illustrate the constitution
of a search domain.
(13)
Mein Auto parkte in der Bismarckstrae.

'My car was parked in the Bismarckstrae.'
The semantic analysis of sentence (13) computes a static localization

fact locating an object, here the car, in a spatial region SR described by
the RO Bismarckstrae and the spatial relation in. Additional pre-processing is not necessary because the semantic fact has the canonical form
required (see 3.2.).
The first step of the conceptual interpretation is the propositional
definition of the primary area. As the distance concept, inclusion is chosen
according to the spatial relation given by the region SR. The object
concept of relevant parts is used to determine a RO-conceptualization
which allows the containment of a car. Such a conceptualization is
provided by the silhouette, which is a conceptualization of the street as
a whole including the buildings on the side of the street and the hollow
space between them. This definition of the primary area is interpreted by
the depictorial system by evaluating the RO-conceptualization and the
distance concept. The result is an area which corresponds to the space
occupied by the whole street (see Figure 6).
Each depiction of a localization is based on the picture of the RO,
which is the basic depiction of the relevant category of objects modified
according to individual features if necessary. For the depiction in Figure
6 the RO is given by a detail of a street in a 2-D projection, the bird's
eye view. The street consists of the lane in the middle and the two sides
including sidewalks, houses, and so on. The depiction of the primary
area is the striped object overlying the picture of the street.
462
Simone Pribbenow
street (detail)
Figure 6. Depiction for example (13)
The second step of the area constituting process is the computation of

the priorizations. Propositional world knowledge contains the information that moving or parking cars are normally located in lanes or
in parking spaces (and not on sidewalks or inside houses). For example
(13), the positive priorization given by the typical location of cars is
assumed to be the only available priorization information. This fact is
passed to the depictorial subsystem, which creates the dotted priorization
area. Now the two localization areas can be combined that means the
operator of definition (12) is interpreted by computing their intersection. 14 The intersection marked by the striped-dotted part of Figure 6
forms the final search domain.
4.2. Influence of contextual information
In the following, I want to discuss the steps of the computation where
meaning representations that are identical up to now can be distinguished
according to different contextual influences. The distinction can take
place at three different steps in the area constituting process:
A. the propositional definition of an area which can evaluate the same
spatial region by different combinations of RO-conceptualization and
distance concept,
B. the depictorial evaluation of the definition which can lead to different
areas, and
C. the computation of priorizations where different spatial contexts can
lead to varying priorization areas and search domains.
463
A. Prepositional definition
For some localization expressions, the definition of the primary area by
a combination of concepts only depends on the information given by the
spatial region of the semantic fact, the reference object RO and the spatial
relation REL. One example are localizations involving the preposition
bei 'near', which are always mapped to the combination of the distance
concept nearness and the neutral conceptualization of the RO.
A non-trivial selection of concepts is necessary for two kinds of localizations:
localization with underlying functional aspects and
localization involving relations of the secondary deixis like vor 'in
front' or links 'left'.
The following expressions (14) show an example of the influence of
functional concepts.
(14)
a. das Kleid im Schrank

'the dress in the wardrobe'
b. die Kleiderstange im Schrank
'the coat rail in the wardrobe'
In both localizations the distance concept of inclusion is selected, but

the conceptualization of the wardrobe varies. In example (14 a) the
wardrobe must function as a container for the LE, the dress. This
functionality can only be supplied by the wardrobe if the dress is included
in the space between the six sides of the RO. Therefore the conceptualization of hollow spaces is used to refer to the space associated with the
wardrobe which is delimited by its sides. The coat rail as LE of (14 b) is
assumed to be a part of the wardrobe itself. A part-whole relation between
RO and LE is provided where the RO excluding the LE serves as
a container for the LE. So the silhouette as the conceptualization pertaining to an object as a whole (see example (13)) must be used.
There are also situations where more than one combination of concepts
is adequate (see 14 c).
(14)
c. der Nagel im Schrank

'the nail in the wardrobe'
The nail could be lying inside the wardrobe, e. g., having been left there
by the people who built it, or it could be a part of it, e. g., holding a
door of the wardrobe in its position. The first alternative corresponds to
the situation in (14 a) and leads to the same conceptualization of the
464
Simone Pribbenow
wardrobe; the second alternative to situation (14 b) and the concepts

selected there. Sometimes the verb involved can be used to solve the
problem: If the nail is lying in the wardrobe than the first alternative is
accepted; if it is driven in the second one is chosen.
The second alternative of a non-trivial selection of concepts are localization expressions involving prepositions like vor 'in front' or links 'left',
which are called prepositions of secondary deixis. These prepositions
invoke the distance concept of nearness and the orientation of the RO
via reference systems corresponding to the specific spatial relation. The
problem to be solved is whether the orientation of the RO must be
computed with respect to the intrinsic or the deictic reference system. If
the RO does not provide an intrinsic reference system then the deictic
system is chosen. In all other cases the problem is not easy to solve as
the different, sometimes inconsistent theories about this selection illustrate. 15
B. Depictorial evaluation of concepts
The second step in making distinctions is the depictorial evaluation of
the concept combination defining the primary area. The interpretation
of conceptualizations based on an extrinsic reference systems and of most
of the distance concepts are influenced by the objects involved in the
situational context.
(15)
a. der Tisch beim Schrank

'the table near the wardrobe'
b. der Papierkorb beim Schrank
'the waste basket near the wardrobe'
For the two expressions in (15), the same propositional definition is

computed: the combination of nearness and neutral conceptualization of
the RO. This definition is passed to the depictorial subsystem to evaluate
the definition. Because of the neutral conceptualization of the RO the
distance concept of nearness is applied to the wardrobe as a whole.
According to the meaning of nearness, the evaluation process has to
select that part of the surroundings where the RO can serve as landmark
for the LE as primary area. As Figure 7 shows, the adequate part of the
surroundings is substantially bigger for the table (Figure 7 a) than for the
much smaller waste basket (Figure 7 b).
On the propositional level, such regularities can only be expressed by
a rule of thumb like "the bigger the object the bigger the area". This rule
a: table near the wardrobe
465
b: waste basket near the wardrobe
Figure 7. The primary areas created for example (15)
applies to reference objects as well as to objects to be located, because

the greater the size of an object, the better it can serve as a landmark or
the easier it can be found. Similar rules are evaluated for the relevance
or the visual salience of the RO. But, as discussed in section 4.1. for
proportional definitions, these rules of thumb are not precise enough to
determine the concrete extent of an area. This can only be done by
depictorial means.
An analogous argument applies to the influence of the spatial context
that also determines the delimitation of areas. Normally, a wardrobe
stands with its back to one wall of the room. This wall functions as a
barrier for the process creating the area. As the two depictions in Figure
7 illustrate for the situation described in example (15), the wall restricts
the area to that part of the surroundings that is inside the room in which
the wardrobe stands. It would not be useful to take the whole surroundings because if the LE is in another room, it cannot be located with
respect to the wardrobe any longer.
C. Priorizations
Propositional means are used to describe the priorizations that are possible in a given situation and how they work. For example, typical
localizations of the LE and typical relations between RO and LE provide
positive priorizations. The resulting search domain is the intersection of
the priorization area and the present localization area, intially the primary
area. Objects competing with the reference object cause negative priorizations that restrict the present localization area (see the example above).
But whether a priorization is applicable for a concrete localization and
466
Simone Pribbenow
what the resulting search domain exactly looks like depend on the spatial
context of the localization.
As an illustration I will discuss example (16) of a negative priorization.
(16)
Mein Auto parkte beim Denkmal.

'My car was parked at the statue.'
Figure 8 a shows the depiction of the primary area around the statue
serving as RO. The priorization to be considered is the negative priorization provided by competing objects. A competing object CO is an
object that can be compared to the R O with respect to size and relevance
and therefore could serve as reference object itself. An LE which is nearer
to the CO than to the original RO normally would be located with respect
to the competing object. So the nearness-area of the C O forms a priorization area restricting the present localization area.
In example (16), the passage whose location is known by the individual
world knowledge of the system is a good candidate for a competing
object. The Figures 8 a - 8 c depict the process of computing the resulting
search domain based on the primary area as present localization area
and the priorization area provided by the nearness-area of the passage
serving as C O shown in Figure 8 b.
a: primary area
b: competing area
c: search domain
Figure 8. The construction of a search domain considering competing objects
In general, which part of the primary area gains a negative priority,

that means loses priority, depends on the competing objects available
and their location. For example, if a C O is too far away from the RO,
so that its priorization area does not overlap with the present localization
area, it will have no influence and will not lead to a specification of this
area. So a concrete priorization with respect to competing objects varies
with the spatial relation between CO and RO; in other words, with the
467
spatial context of the considered localization. It should be noted that in

the system presented, this influence of the spatial context can only be
computed by the depictorial module.
5. Concluding remarks
Most parts of the process explained in the Sections 3.2. and 4. were
implemented by the LILOG-space project at the university of Hamburg.
As stated in 2.1., the processing of localization expressions is embedded
in the LEU/2 system, which is a prototype of a text understanding system
developed by IBM Germany in co-operation with five universities. The
propositional part of conceptual processing is represented in the form of
rules using a kind of order-sorted predicate logic language developed in
the LILOG project (see Pribbenow to appear or 1991 for an overview of
the rule-based processing). The depictorial processing is based on cell
matrices as a realization of depictions (see Khenkhar 1991).
In the future, it will be interesting to investigate the extent to which
the conceptual basis for localization and its processing can be used for
other tasks besides the analysis of localization expressions in a text
understanding system. Candidates for other domains are geographical
information systems, e.g., natural language interfaces to such systems,
or in the more distinct future computer vision or naive kinematics,
the spatial aspects of physical commonsense reasoning.
Notes
1. The first object is called "entity to be located" and not "object to be located" because
there is also the possibility to locate situations and motions as the examples (3) and
(4) will show.
2. The third section will show why it is necessary to assume the localization of situations
on the semantic level and how the localization of situations can be reduced to that of
certain objects involved in the situation under consideration.
3. Another approach is given by Talmy (1983). Often his conceptual basis overlaps with
the one presented in this paper, but some aspects such as the distance concepts are
missing in Talmy's approach.
4. Talmy and Herskovits assume that the RO and the LE interact with each other according
to the figure-ground principle of gestalt theory. The LE which they call figure is seen
as isolated from the ground provided by the RO. But the experiments of Dittrich
Herrmann (1990) suggest that the figure-ground principle is too restricted to explain
468
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
Simone Pribbenow
the roles of RO and LE in all their aspects. So my general localization principle does
not restrict the LE and the RO to pure figure/ground entities.
According to Lynch (1960), a landmark can be provided by any object that is used for
orientation tasks, for example a building or a crossing.
Slightly different classifications of reference systems given by other authors can be
found in Retz-Schmidt (1988).
The term "distance" may be not the best choice because two of the distance concepts
(inclusion and contact/direct nearness) describe the lack of distance.
The term "search domain" is taken from Miller Johnson-Laird (1976) who use the
term in an abstract way, not only in spatial contexts but for all concrete or conceptual
spaces that are searched for a special object.
The linguistic and the conceptual level of the process in this figure correspond to the
processing modules of the same name in Figure 1. The two transformation steps in
Figure 5 are carried out by the propositional subsystem of the conceptual processing,
the area constituting process by both the propositional and the depictorial subsystem
of the system in Figure 1.
For a complete description of the linguistic analysis see Maienborn (1990) or Pribbenow
(to appear: 2.3).
Unfortunately, the process of transforming dynamic localization facts is more complicated than it seems according to the brief description given above. The process is
explained in detail in Pribbenow (to appear: 2.4).
Remember that all other object concepts (form, dimensionality and plurality) are used
for requirements to LE and RO. These requirements are used to reject given localization
expressions and restrict the generation of new ones.
The evaluation of a RO-conceptualization focuses only on parts of the object without
forgetting the rest of it. This process is not a selection of pieces.
The intersection of areas is one possible interpretation of the operator
As example
(16) will show, a different evaluation is needed for negative priorizations evocated by
competing objects.
Compare Retz-Schmidt (1988) for a collection of some of these theories.
References
Andre, Elisabeth Gerd Herzog Thomas Rist
1988
"On the simultaneous interpretation of real world image sequences and their
natural language description: The system SOCCER", in Proceedings of the
ECA1-88, 449-454.
Bierwisch, Manfred
1983
"Semantische und konzeptuelle Reprsentation lexikalischer Einheiten", in:
Rudolph Ruzicka Wolfgang Mtsch (eds.), 61-99.
1988
"On the grammar of local prepositions", in: Manfred Bierwisch Wolfgang
Mtsch Ilse Zimmermann (eds.), 1-65.
1987
Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven. Berlin:
Akademie-Verlag.
469
Bierwisch, Manfred Wolfgang Mtsch Ilse Zimmermann (eds.)

1988
Syntax, Semantik und Lexikon. (Studia grammatica 29.) Berlin: AkademieVerlag.
Brachman, Ronald J. Victoria Gilbert Hector J. Levesque
1985
"An essential hybrid reasoning system", in: Proceedings of the 9th IJCAI,
532-539.
Davis, Ernest
1990
Representations of commonsense knowledge. San Mateo, Cal.: Morgan Kaufmann.
Dittrich, Sabine Hermann, Theo
1990
"Der Dom steht hinter dem Fahrrad." Intendiertes Objekt oder Relatum?.
(Bericht Nr. 16.) Heidelberg/Mannheim: Arbeiten aus dem SFB 245
"Sprechen und Sprachverstehen im sozialen Kontext".
Freksa, Christian Christopher Habel (eds.)
1990
Reprsentation und Verarbeitung rumlichen Wissens. Berlin: Springer.
Garrod, Simon C. Anthony J. Sanford
1988
"Discourse models as interfaces between language and the spatial world",
Journal of Semantics 6: 147-160.
Habel, Christopher
1987
"Cognitive linguistics: The processing of spatial concepts", in: T.A. Informations ATA LA (Association pur le Traitement Automatique des Langues)
28: 21-56.
Habel, Christopher Pribbenow, Simone
1988
Gebietskonstituierende Prozesse. (LILOG-Report 18.) Stuttgart: IBM.
Herskovits, Annette
1986
in English. Cambridge: Cambridge UR
Herweg, Michael
1990
Zeitaspekte. Die Bedeutung von Tempus, Aspekt und temporalen Konjunktionen.
Wiesbaden: Deutscher Universitts-Verlag.
Herzog, Otthein Claus-Rainer Rollinger (eds.)
1991
Text understanding in LI LOG: Integrating computational linguistics and artificial intelligence. Berlin: Springer.
Hobbs, Jerry R. Moore, Robert C. (eds.)
1985
Formal theories of the commonsense world. Norwood, N. J.: Ablex Publishing
Corporation.
Johnson-Laird, Philip N.
1983
Mental models. Cambridge: Cambridge UP.
Khenkhar, Mohammed
1991
"Object-oriented representation of depictions on the basis of cell matrices",
in: Otthein Herzog Claus-Rainer Rollinger (eds.), 645-658.
Klein, Wolfgang
1990
Raumausdrcke. [Unpublished MS, Nijmegen.]
Kosslyn, Stephen M.
1980
Image and mind. Cambridge, Mass.: Harvard UP.
Lindsay, Robert K.
1988
"Images and inference", Cognition 29: 229-250.
470
Simone Pribbenow
Lynch, Kevin
1960
The images of the city. Cambridge, Mass.: MIT-Press.
Maienborn, Claudia
1990
Lokale Verben und Prpositionen: Semantische und konzeptuelle Verarbeitung
in LEU 12. (IWBS-Report 119.) Stuttgart: IBM.
1976
Language and perception. Cambridge: Cambridge University Press.
Paivio, Allen
1983
"The empirical case for dual coding", in: Yuille, J. (ed.), 307-332.
Pick, Herbert Linda Acredolo (eds.)
1983
Spatial orientation: Theory, research, and application. New York: Plenum
Press.
Pribbenow, Simone
1990
"Interaktion von propositionalen und bildhaften Reprsentationen", in:
Christian Freksa Christopher Habel (eds.), 156-174.
1991a
"Phenomena of localization", in: Otthein Herzog Claus-Rainer Rollinger
(eds.), 609-620.
to appear Rumliche Konzepte in Wissens- und Sprachverarbeitung. Hybride Verarbeitung
von Lokalisierung. Deutscher Universitts Verlag.
1988
"Various views on spatial prepositions", AI Magazine 9/2: 95-105.
Ruzicka, Rudolph Wolfgang Mtsch (eds.)
1983
Untersuchungen zur Semantik. (Studia grammatica 22.) Berlin: AkademieVerlag.
Schirra, Jrg
1990
"A contribution to reference semantics of spatial prepositions: The visualization problem and its solution in VITRA", in this volume.
Stechow von, Arnim Dieter Wunderlich (eds.)
1991
Handbuch der Semantik. Berlin: de Gruyter.
Talmy, Leonard
1983
"How language structures space", in: Herbert Pick Linda Acredolo (eds.),
225-282.
Vandeloise, Claude
1985
Descriptions of space in French. (Paper A 150.) University of Duisburg:
Linguistic Agency.
Vieu, Laure
1991
Semantique des relations spatiales et inferences spatio-temporelles: Une contribution l'etude des structures formelles de l'espace en Langage Naturel.
[Docteur these, L'Universite Paul Sabatier de Toulouse.]
Wunderlich, Dieter Herweg, Michael
1991
"Lokale und Direktionale", in: Arnim von Stechow Dieter Wunderlich
(eds.)
Yuille, J. (ed.)
1983
Imagery, memory and cognition. Hillsdale, N.J.: Erlbaum.
A contribution to reference semantics of spatial

prepositions:
The visualization problem and its solution in
VITRA
Jrg R. J. Schirr a
1. On reference semantics in AI
In AI research concerning natural language systems, the reference aspect
of verbal expressions very often plays only a minor role. Nevertheless
one has to consider that every verbal expression refers to something, and
that the structure of this something has an influence on the use of that
verbal expression, i. e., on its meaning. In other words: the meaning of
any verbal expression is somehow anchored in the corresponding referents. This extra-linguistic influence is especially recognizable if we study
the interactions of a natural language system with the world by means
of sensor and motor systems, e. g., a vision system. But also in machine
translation, considering the referents will help to overcome the gap
between the conceptual systems of two languages which has so often
trapped even approaches with a relatively deep semantic analysis of the
source texts (cf. Zelinsky-Wibbelt 1990 a, 1990 b; Bateman 1990, and
Grabski 1990). The (more or less) unique referent might serve as a fixed
point during the transformation to the goal language: changes in the
meaning structure necessary due to the different conceptual systems
underlying the two languages are restricted to valid interpretations of
that referent.
The major question arising in this context is: What is the nature of the
reference relationship? which can be split into (a): What are the referents?
and (b): How does their influence on verbal behavior work? I will
concentrate herein on a primitive form of spatial prepositions and their
referents essentially geometric relations although some of these
considerations are more general (cf. Schirra 1990 b).
Analyses of spatial prepositions in the framework of reference semantics usually view geometric relations as in some sense objectively given,
i. e., existing independently in the so-called "real world", external to any
472
Jrg R. J. Schirra
mind (e.g., cf. Dowty et al. 1981). However, a more careful examination
leads to the conclusion that the needed referents cannot be provided by
the world per se. Expressions of fictitious things show us one argument
against this objectivist view of the reference relationship: there are no
unicorns in the real world, hence no objectively given referents for the
expression "The last unicorn went back into the silent forest". Is it possible
that we can understand this expression even without referents?
There are other simpler disadvantages of the objectivist view. Imagine
that we speak about a journey around the world we want to take in the
future. While standing in the middle of Europe, we might use a sentence
like: "Afterwards, we will fly to New Zealand which is to the east of
Australia." Here, we do not have the problem of nonexistent referents:
"to the east" clearly refers to that de facto spatial relation we could
perceive if only we would look for example from a spaceship at that part
of the earth. But how can this referent have any influence on the dialog,
i. e., on the use or meaning of verbal expressions, miles away on the other
side of the earth? Thus, not only is the nature of the referents in some
cases unclear to say the least (cf. Figure 1). Also our second question
Figure I. How can the reference relationship overcome the distance?
Spatial prepositions: The visualization problem
473
about the kind of influence referents have on verbal behavior remains

essentially unanswered (cf. note 1).
The alternative experiential view of the reference relationship assumes
that referents are essentially percepts and therefore always mental constructs (cf. Lakoff 1987 and Johnson 1987). Thus, spatial prepositions
refer mainly to visually perceptible relations between objects. Now, there
are no obvious problems concerning the mechanisms of influence, as long
as we only speak about perceived things. But the examples above still
remain problematic. How can the spatial relation between Australia and
New Zealand influence a dialog without being perceived? We seemingly
need some kind of pseudo-percepts, if we speak about something not
present or fictitious. It is this very nature of percepts as being mentally
constructed which allows for constructing other mental entities and using
them as substitutes for percepts. Obviously, we have to speak about these
hypothetical mental entities in just the same manner as about percepts
which is exactly the way we speak about mental images (cf. note 2).
Thus, if we consider the reference relationship experientially rather than
from an objectivist point of view, it is possible to better understand visual
mental images and their cognitive function: mental images have the
cognitive function of making available visual referents in the case where,
as for radio reporting of a sports event, they are not directly perceptible
(cf. Figure 2).
Figure 2. Mental images as substitutes for percepts
Dealing with reference semantics in AI, one has to comprehend that

the reference relationship consists of three parts: (a) the referents; (b) the
474
Jrg R. J. Schirra
propositions, which are the mental representatives of the meaning of

verbal expressions; and (c) the connection between the two, the reference
relation proper, so to speak (cf. note 3). These three entities all have to
be modeled in an AI system in the framework of reference semantics (cf.
Figure 3).
referential level
the
reference relation
proper
propositional level
Figure 3. The reference relationship a first view
The propositional level is already well examined in AI: most formalisms

of Knowledge Representation can be used for this purpose. On the other
hand, the referent level because of its connection to perception has
to be based on data structures used in corresponding perceptual systems,
e.g., vision systems (e.g., cf. Marr 1982; Sung 1988; Herzog et al. 1989,
and Mohnhaupt and Fleet 1988). At this point, one confusion of notions
very often appears, blurring further discussions: since the results of
perceptual systems in AI even on a very low level of processing are usually
represented in a form similar to the above-mentioned formalisms used
for the propositional level sometimes even the very same knowledge
representation languages are used they are called propositional, as
well. This use of the word "propositional" is different from the one
introduced above. In this very broad sense, everything expressed in the
knowledge representation formalism is called a proposition (cf. Pylyshyn
1981; note 4). In my terminology, only those expressions which are
connected to referents by a reference relation are propositions. The
difference between referents and propositions is not a question of form
but of use. Referents have meaning, namely the propositions associated
with them, but they do not refer to something else, e. g., an extra-mental
set-theoretical model named "the world". Similarly, propositions do not
have meaning; they only refer to something. There is neither a meta-level
above the propositions which might include the meanings of propositions,
nor is there a sub-level which could keep ready those objects the referents
refer to (cf. note 5).
475
The third and most important entity of the trinity of reference, the
connection between referents and propositions, quite obviously has a
very different character. In order to avoid an infinite regress, the realization of the reference relation proper cannot use propositions or referents
(percepts) again. Actually, the system does not necessarily know anything
about the structure of the reference relation, e. g., for a spatial proposition.
It can only use its realization of the reference relation in order to establish
the connection between referent and proposition, i. e., recognize spatial
relations in a given percept or visualize spatial relations (construct the
referent). Therefore, a procedural realization of the reference relation
proper seems adequate.
In the following sections, I will present an overview of the system
SOCCER with special emphasis on how language can be grounded in
visual perception using a simple kind of artificial percepts, and how
mental images of the same simple kind can be reconstructed in a model
of anticipated understanding. The reference relation is used for, and its
realization restricted by, both purposes. Section 6, finally, presents the
realization of the reference relation we have chosen in VITRA.
2. The project VITRA and the system SOCCER

The project VITRA (Visual TRAnslator) which started in 1985 as part
of the German special collaboration programme SFB 314, AI & Knowledge-Based Systems, examines the relations between speaking and seeing:
a completely operational form of reference semantics for the visually
perceived is to be developed. CITYTOUR and SOCCER are two systems
constructed in VITRA which broadly speaking transform visual
perceptions into language. Here, we will concentrate on SOCCER (for
CITYTOUR cf. Andre et al. 1985, 1986 a, 1986 b; Schirra et al. 1987,
and Retz-Schmidt 1988).
SOCCER simultaneously analyses and describes in German short
scenes from soccer games similar to a live radio report, i. e., simultaneously and in an objective manner to an audience which is not able to see
the game themselves. For this purpose, a large number of quite complicated 'cognitive' activities has to be performed: e. g., perceiving the
locations and movements of ball and players in the scenery, interpreting
these movements with respect to the conventions of soccer games, especially assuming the intentions and plans of the agents in the field, and
476
Jrg R. J. Schirra
last but not least selecting which events to utter in which sequence
and with which words.
The input data of SOCCER, which in a way are its percepts, are called
mobile object data MOD and generated by the motion analysis
system ACTIONS (cf. Nagel 1988; Sung 1988; and Herzog et al. 1989).
These percepts consist of the set of the two-dimensional spatial locations
and the velocity vectors of every mobile object perceived in the soccer
field from a bird's eve view. At every time quantum, ACTIONS delivers
the corresponding data which, then, is entered successively into SOCCER.
At present, all mobile objects are perceived as ideal points (zero-dimensional). The Mobile Object Data is analysed successively as soon as it is
entered. It implicitly refers to the geometry of the soccer field which is
known by the system as StaB static background (cf. Figure 4).
MIN
SEC
1/100
D*l*y: 2
SeH: 1.9
Bee*: : :
End: B8:2B:80
: :1:
1 I n k e r : O f t On
S h o u F o c u : No V*
TRACE: Orr On
R o r c H n B ! Oll On
OUTPUT

i * k l , 4r V t r t U p p r , l i t
Ball,
w a i u r , *r K k a u m h , Ut U i f t i i t o d u .
Er g r e i n U M , V m u r p t r , .
Dar A W * k r * f U i r k i t r w i t r e a .
Dar A b w k r f f l l r k W i e a A*xrUft*pl)*r i i | ( ( r l ( l i .
Dar A K g r i i l M f U l a r ) f U l l Mickali, tal V a r U a r , 4a Ball
tat m a
*
C i t IMM
Dt MO
fat CLOCK
T < r i < l l u l i n MW.
CnhiuI k|tiH
u j
1 r 1 < (J 111 ,111(1 M i d , I

u c 2 f Nov B : S 9 : 2 3 J S c h T f f r e r v i
tieep t i g h t .
Gaehn
Figure 4. What SOCCER shows the user
SOCCER does not know the whole scene at once. Like a radio reporter,
it has to analyse the scene during its occurrence. Therefore, all processing
steps have to be done incrementally, i. e., a selection of already recognized
events is verbalized simultaneously with further event recognition. Indeed,
SOCCER already recognizes events before they have occurred completely.
477
Scene
Figure 5. SOCCER: The core system
The power, but also the limitation of a rigorous grounding of language

in perception is already demonstrated by the core system (cf. Figure 5):
in a kind of pipelining, three components transform the perceptual data
into a text similar to the protocol sentences (as they are called) of Vienna
Circle philosophy (cf. Carnap 1933 and Neurath 1933): only directly
sensed impressions are reported indeed still too primitive a type of
sports report. For example, compare text (a) with (b), both of which
describe the same referent.
a. Schmidt, the goal keeper of the Blacks, is standing in the left penalty
area. Meier, the captain of the Reds, is running along the middle line.
The ball is close to him. Now, the distance between the ball and Meier
increases. The ball crosses the left half of field and comes near to
Schmidt. Schmidt starts moving towards the ball. Now, the ball stops
moving when it reaches Schmidt.
478
Jrg R. J. Schirra
b. Meier, the captain of the Reds, has the ball and is running along the
middle line. Now, he tries to score with a long shoot, but Schmidt, the
goal keeper of the Blacks, catches the ball in his penalty area.
The first component of the core system, S O C C E R ' s Event Recognition,
has two parts: first, elementary spatio-temporal relations are recognized
in the current percept: propositions like (left player-5 player-7) or (greatervelocity ball 45) are constructed by algorithms which are procedural to
S O C C E R . The core of these algorithms is always a graded classification
function that associates visual percepts used as input to S O C C E R with
abstract spatial relations by so-called applicability degrees (cf. Section 4).
As a second step, these elementary relations are chunked into propositions describing more abstract relations: (running-with-the-ball agent:
player-5 place: (in-front-of RightGoal)) or (running-parallel agent: player7 co-agent: player-2 direction: (along MiddleLine)). In contrast to the
elementary relations, S O C C E R here uses knowledge in a declarative form:
for each composed event, a so-called event model defines how elementary
relations have to be combined to yield events of that type. The event
recognition component works in a quasi-parallel way, thus recognizing
simultaneously all spatio-temporal relations or events similar to a h u m a n
observer ( S O C C E R event recognition: cf. Herzog and Rist 1988).
While the event recognition is working, the Selection component already chooses subsets of the recognized propositions to be uttered and
orders them into a queue. The head of this queue is passed to the
Generation component as soon as it is idle. The time of generating the
sentences is important since the order in the queue might be changed if
new events which seem to be more relevant have been recognized in the
meantime. Previously selected items might even be removed from the
queue.
The generation component transforms the event proposition chosen
by the selection component into a continuation of the report. Here, the
applicability degrees of elementary spatio-temporal relations are verbalized, using linguistic hedges (cf. Figure 6 and Lakoff 1972). Localizing
phrases and even previously mentioned events can be used to disambiguate objects. Furthermore, anaphora are used to increase the coherence
of the text (a detailed description of both selection and generation components is to be found in Andre 1988).
Obviously, the S O C C E R core system is rather primitive compared to
h u m a n perception and cognition: it "sees" only two spatial dimensions,
namely soccer fields from a bird's eye view, and "perceives" players as
479
appr
behi
more or
behind
directly behind
Figure 6. Applicability degrees and linguistic hedges
zero-dimensional without inherent orientations. Correspondingly, its language use is restricted: spatial prepositions for example cannot be used
in the differentiated way we use them. Especially, there are no metaphoric
extensions of spatial prepositions. On the other hand, these limitations
simplify the problems of realizing the reference relationship of spatial
prepositions to a treatable complexity, and thus, even may serve as a
base for further studying spatial metaphors.
To overcome the handicaps of pure protocol sentences, the intention
and plan recognition component REPLAI-II (cf. Retz-Schmidt 1991b)
is added to the SOCCER core system (cf. Figure 7): we can actually
speak about soccer games only if we assume that the players behave
according to internal and not directly observable mental states like
intentions and plans. If the ball moves into the goal after changing its
direction while being very close to a player, SOCCER can only describe
this event as "That player scored" if it presumes and tries to verify in its
percepts that this player has certain soccer-specific intentions and plans.
Although intentions are based in the perceived, as well, they are not
totally grounded in percepts and require additional knowledge to be
(hypothetically) recognized. Thus, REPLAI-II extends the strict reference
semantics of the SOCCER core system (cf. note 6).
Another extension of SOCCER is provided by its listener model.
480
Jrg R. J. Schirra
Scene
Figure 7. Extended architecture of SOCCER
3. The listener model ANTLIMA

In order to follow Grice's Cooperative Principle ("Make your conversational contribution such as is required, at the stage at which it occurs,
by the accepted purpose or direction of the talk exchange in which you
are engaged." Grice 1974) a speaker has to know how his utterance is
understood by the listeners in the present context. He needs a model of
the listener, e. g., to make sure that despite the maxim of economy ("Make
your contribution as informative as is required [for the current purposes
of the exchange]." Grice 1974) the listeners are still able to recognize all
the relevant structures even from elliptic descriptions. Thus, a listener
model serves as a device which balances between the divergent demands
of economy and completeness ("Do not make your contribution more
informative than is required" Grice, 1974). With this knowledge about
the listeners, the speaker can rate how much information actually is
required in the given case.
481
Correspondingly, SOCCER also needs a component that can construct

a model of the listeners' knowledge of the events that have already been
described. This listener model enables the system to continue its description in a cooperative way by anticipating the listeners' understanding of
the utterance just planned. With these anticipations, the plausibility of
that utterance in the context already known can be rated and used in an
anticipation feedback loop to improve the coherence (cf. Figure 7).
Therefore, the main task is to find out whether and to what degree the
listeners might be able to understand the planned utterance at all, and,
as a second step, whether they understand it as intended (cf. note 7).
There is little evidence that the listeners and the speaker use different
kinds of semantics. Thus, we assume that the listeners understand the
soccer report by reference semantics, as well. What does that mean? As
a German linguist wrote in 1969, "The radio reporter has solved his task
only if he describes the reality of a sports event so vividly and obviously
to the listener that the listener believes he sees that reality." (Dankert
1969: 94 [transl. J. S.]) The reporter shall induce so to speak a
cinema in the heads of his audience. This clearly refers to mental images
just in the way we mentioned them in Section 1: if the listeners want to
have a "deep" understanding of the report, they need access to the
referents and should be able to reconstruct them, i. e., to construct (visual)
mental images corresponding to the speaker's percepts (cf. note 8).
Since the listener model of SOCCER anticipates how the listeners
understand the planned continuation of the report, it also has to ground
the meaning of these utterances and especially of spatial prepositions
referentially: ANTLIMA ANTicipation of the Listeners' IMAgery
must be able to reconstruct corresponding visual percepts albeit in the
limited sense of SOCCER as Mobile Object Data and Static Background
(cf. Figure 8). In other words, it must visualize the abstractly described
situation. It is our thesis that those pseudo-percepts reconstructed by
SOCCER's listener model correspond to the listeners' visual mental
images (cf. note 9).
As the backbone of ANTLIMA, the event proposition chosen by the
selection component of SOCCER has to be analysed conceptually, i.e.,
dealing only with propositions. The conceptual analysis provides the basis
for the visualization and for the feedback to SOCCER: the definition of
the considered event, i. e., its subevents and the spatio-temporal relations
between them, must be expanded and adapted to the situational constraints in order to achieve spatio-temporal coherence with the context
(cf. Schirra 1990 b). Additionally, it might be necessary to integrate
482
Jrg R. J. Schirra
Figure 8. Architecture of A N T L I M A
modifications given by optional deep case fillers (cf. Sondheimer 1978

and Marburger and Wahlster 1983). The construction of a corresponding
mental image serves as a kind of focusing device for this reasoning (cf.
Pribbenow 1988 and Schirra 1991, but also Schirra 1992 b). The intermediate result of the conceptual analysis, which is the basis for the
visualization, is called the propositional elementary structure, the temporally ordered sequence of sets of elementary spatio-temporal relations.
Visualization transforms this data to a sequence of Mobile Object Data
in the Static Background, thus fixing particular locations and velocities
for all considered objects at each time point.
The generated mental image should not be compared directly with the
original percept of SOCCER coordinate by coordinate, so to speak.
Although such a comparison seems to be necessary if we want to know
whether the listeners will have got the correct referent, the system only
compares propositions, not images. Why? In general, we lose information
transforming an image to propositions. Therefore, we cannot expect that
the listeners will reconstruct the very same picture from that selected set
of propositions actually communicated. Their images and equally the
one generated by ANTLIMA will usually be only more or less similar
to the original percept. The question now is: which deviations are essential
and which are not? If a player is standing in the middle of the right field
nobody near him 50 pixels' difference normally will not matter.
But if he stands at the edge of the field, or near some other player, or
very close to the ball, even 10 pixels' deviation of his location might
change the whole interpretation of the scene. Since the propositions just
483
ignore irrelevant details by definition (cf. Section 4), A N T L I M A has to

re-analyse its mental images first, i. e., describe them propositionally again
(cf. Figure 8, component Re-analysis). Then, it can compare this new set
of propositions with the set SOCCER "found" in the original percept or
the set which was actually communicated. Now, substantial deviations
result in a different set of propositions, e. g., an additional (outside player5 SoccerField) or a missing (at player-7 ball). Since these differences
might have several causes, among them: (a) error of A N T L I M A during
conceptual analysis or visualization; (b) error of SOCCER during selection; (c) expectation of an event which will be communicated next, and
(d) implicit event which SOCCER expects A N T L I M A to know about
without communicating it, the further processing is quite complicated.
As it does not effect the discussion in this report, we do not deal with it
herein (cf. Schirra 1991). Finally, plausibility ratings, correctly expected
continuations of the report, and if necessary errors are given back
to SOCCER's selection component, thus closing the anticipation feedback
loop.
Before the elementary visualization is described in greater detail in
Section 5, we devote our attention to its counterpart, the recognition of
elementary spatio-temporal relations, since this first use of the reference
relation gives us a clue for the solution of the visualization problem.
4. Recognition of static spatial relations

Dealing with the connection between seeing and speaking, the first of the
problems we have to consider is establishing the reference relation at all:
we have the task of finding a connection between perceptual and propositional level. In general, the transformation to the propositional level
has the function of reducing the amount of information included in the
percepts to those features important for further acting reporting in
our case. Therefore, the spatial relations we consider correspond closely
to (German) prepositions. The location of an object is relevant only
relative to other objects' positions. The first object is usually called LO
located object the others ROs reference objects (cf. note 10).
Other information in the percepts, e. g., the precise coordinates of objects,
is regarded as irrelevant, and ignored.
The elementary level of recognition in SOCCER is formed by static
spatial relations verbally described by prepositions such as "being left
484
Jrg R. J. Schirra
o f ' , " at", and " between" (actually, of course, in German) and
represented by means of what I call "Spatial Concepts". Individual
occurrences of a relation are represented by instances of the Spatial
Concept and called "(spatial) propositions". More precisely, spatial propositions are combinations of one spatial relation the type of the
proposition and a set of objects forming the arguments of the relation.
Figure 9 shows a typical (static) percept for SOCCER. Corresponding to
Figure 9. SOCCER Percept with near and left of occurrences
the human uses of spatial prepositions, SOCCER should be able to

interpret this percept as a "near" situation, i.e., by creating a proposition
(near player-5 player-7). But it can instantiate the Spatial Concept "lefto f ' with respect to the direction of movement, as well. Obviously,
recognition is not a one-to-one association, i. e., simply combining one
percept with one proposition (cf. Figure 3). Different propositions can
stand for the same geometric configuration. The reference relationship
in VITRA associates every percept with a set of propositions, all describing that percept (cf. Figure 10). But which of them should be used to
describe the percept verbally?
This decision is supported by the radial structure of spatial concepts:
we have to consider that some referents are good, others bad examples
of a proposition. In most cases, we can move the LO a little bit without
changing the propositional description. But the referents gradually become poorer examples of that proposition. For most spatial relations, we
485
Figure 10. The reference relationship (Part 2: Recognition)
find that the probability of their being used to describe a situation changes
gradually as the LO shifts. Figures 11, 12, and 13 illustrate this phenomenon for "being in"; " near"; and " in front o f ' by means of a kind
of probability clouds, each drawn for two different types of ROs. The
Figure 11. Probability cloud representation of "in" for two different kinds of objects
486
Jrg R. J. Schirra
Figure 12. Probability cloud representation of "near" for two different kinds of objects
dense centers of these clouds mark those positions rated as good examples
for the relations.
In VITRA, the probability of use of a spatial proposition which is
associated with every possible position of the LO with respect to some
given ROs is interpreted as a measure of applicability for the proposition.
In Figure 10, the degree of applicability is indicated by the thickness of
the connection. The higher the applicability degree of a proposition is
for a given percept, the better this proposition can be used to describe
the percept. Therefore, an essential part of the reference relation is
encoded by a graded classification function associated with every proposition. Given a percept, these functions yield the applicability degree
(in [0.0 ... 1.0]) of the corresponding proposition. For example, the
applicability degree of "being at" is calculated by means of Formula 1
only depending on one parameter: "dist", the distance between the LO
and the RO; "d" is a scaling factor.
dist
Aar (dist) =
487
Figure 13. Probability cloud representation of "in front o f ' (extrinsic use) for two different
kinds of objects
The connections between spatial concepts, propositions, percepts, and

applicability degrees are shown in Figure 14 as signatures of abstract
data types (ADT). Note that the reference relation and the abovementioned classification functions are hidden in the operation recognize.
For reasons described in the next section, the applicability degrees are
called "T-values".
Ambiguity and gradation are both well described in linguistic literature:
many investigations about the connection between object locations and
applicable spatial relations have been carried out and can be used for
our purpose (e. g., cf. Saf 1966; Fillmore 1971; Miller and Johnson-Laird
1976; Moilanen 1979; Talmy 1983; Herskovits 1985; Vandeloise 1986;
Lakoff 1987; Bierwisch 1987; and Habel et al. 1989). But, to be sure, the
simplicity of SOCCER's percepts does not allow for the full range of use
German prepositions show: for example, metaphoric uses even in the
spatial domain (cf. Schirra and Hays 1993) are totally excluded. Nevertheless, some of the features described by linguists can be and have been
considered in SOCCER (cf. Sections 6 and 7, and note 11).
488
Jrg R. J. Schirra
Figure 14. Signature of the spatial concept system (I): Recognition
5. Visualization of static spatial relations

Corresponding to the elementary recognition, the most elementary step
of the visualization task is constructing the static image of a set of
elementary static spatial propositions which should hold simultaneously
(cf. Section 3 and note 12), e.g.,
(left player-5 RightPenaltyArea), (in-front player-5 player-7),
(in player-3 RightHalfField), (near player-3 MiddleLine),
(near player-7 OutfieldSidel), (at player-7 player-3),
(near player-7 ball), (between player-7 player-3 player-5),
or organized with respect to the LOs:
player-5: [left RightPenaltyArea], [in-front player-7];
player-7: [between player-3 player-5], [at player-3], [near ball], [near
OutfieldSidel];
player-3: [in RightHalfField], [near MiddleLine]
This task, called the visualization problem in ANTLIMA, is just the
reverse of the aforementioned classification task that is at the heart of
perception: instead of abstracting away from a concrete situation, an
abstract (propositional) description must be augmented, filled with life
that is, and transformed into a plausible concrete form. Unfortunately,
the advantage of classification, namely the possibility of ignoring irrele-
489
vant details of the referent, produces for the visualization task a corresponding disadvantage: how, for example, can we fix a concrete (i.e.,
precise) position of a ball of which we only know that it is to the left of
the penalty area?
One way out of this problem is given by the already mentioned
gradation of the Spatial Concepts: if we consider the reference relationship
not from the viewpoint of a given percept with respect to which every
spatial proposition is applicable to a certain degree, but conversely from
the viewpoint of one such proposition, then several percepts are to varying
degrees typical examples of that proposition. Whereas the problem of
perceiving is essentially the question: Which (spatial) relation is most
applicable to a (visually) given situation?, the visualization problem can
be summarized as: Which situation (visual pseudo percept) is most typically intended by a given abstract description? (cf. Figure 15).
Therefore, I assume that listeners and similarly A N T L I M A

always try to give an utterance its most typical interpretation. Furthermore, they expect that the speaker will explicitly mention any important
deviation from the typical case. This again reflects the aforementioned
criterion for deciding which of all applicable propositions should be used
to describe the percept to the listeners: SOCCER chooses the propositions
with the highest degrees of applicability for verbalization.
Starting from a propositional description, that mental image must be
constructed that realizes all of the given spatio-temporal relations with
maximal typicality. For a given set of ROs, A N T L I M A has to locate the
LO at a position where the classification functions, or typicality functions
as they better might be called in this context, are maximal. The essential
490
Jrg R. J. Schirra
task then is finding the maxima of the typicality functions. This task is
not too difficult for one proposition. However, we have to consider sets
of propositions which should hold simultaneously. Some of the given
relations may happen to conflict: they cannot be maximally typical for
the same situation. Look, for example, at Figure 9: if the ball is to be
located in the left goal and simultaneously near player-7, we are not able
to find any location with maximal typicality for both restrictions. In
those cases, compromises must be calculated: a renormalized addition of
all the typicality functions (e.g., algebraic average) which are associated
with each LO meaning in this case: object to be located describes
the typicality distribution for the conjunction of the restrictions. We can
illustrate these calculations as a combination of several probability clouds
(cf. Figures 11 to 13): only where several dense centers overlap does the
combined typicality reach really high values. Figure 16 shows the combination of three localizations: [near CenterCircle], [between player W3
RightGoal], and [at MiddleLine] (cf. note 13).
With such combinations, compatible restrictions result in a combined
typicality function with a quite high typicality maximum, as shown in
0:00.00
mm sec UfOO
,11 Propositions
ANTLIKA command
wm- comma/id
:
. command
paANTLIKA command
Change Parameters
Clear
Configurations
Demo
Events
File
Sequences
Show
Typof Editor
Change Parameters
Change Parameters
Change Parameters
Figure 16. Combination of three applicability clouds with sectional view on the right side
491
Graphic Pang
Figure 17. Combination of applicability clouds of compatible spatial propositions
Figure 17 for the combination of [in front of LeftGoal] and [near player7], whereas incompatible propositions, e. g., [in front of RightGoal] combined with [in front of LeftGoal], yield maxima of typicality with an
extremely low value, since the maxima of the components do not overlap
(cf. Figure 18 with maxima at 0.5). Thus, the maximal degree of typicality
or T-value reachable for the set of propositions can be used as a
rating for the plausibility the considered utterance has for the listeners.
If only a low T-value can be reached while generating the mental image,
i.e., reconstructing the referent of that utterance, incompatibilities must
have been included. If a high typicality value can be reached, all involved
restrictions could be satisfied in the mental image, and the listeners found
(one of) the typical referent(s) of the utterance.
Using again the signatures of abstract data types, we can represent the
visualization as an operation of the same set of ADTs already used in
Section 4. Figure 19 shows that the operation visualize takes an initial
image and a set of propositions and yields another image (cf. note 14).
The initial image is generated by an operation init-image which takes the
492
Jrg R. J. Schirra
Figure 18. Combination of applicability clouds of incompatible spatial propositions
Figure 19. Signature of the spatial concept system (II) Visualization
493
set of all considered objects and locates them at arbitrary positions (cf.
note 15). The special attribute of the resulting image cannot be represented
in the signature. Instead, I use the following logical formula (cf. Formula
2):
2.
V P e 2proposiom, V/ e referent:
X (recognize (visualize (init-image I U
P
(arguments? (/?'))
Ve/>
),
>
peP
))
(recognize (I, ))
Or less formally: if we sum up the T-values (typicality/applicability

values) of all propositions in the given set of propositions with respect
to the result of the visualize operation on P, this sum has to be at least
as great as the sum with respect to any other possible image. In other
words: the result is the (or one of the) most typical image(s) for the set
of propositions P.
6. The operational form of the reference relation in VITRA

In the preceding two sections, the reference relation was used in two
directions, for recognition and for visualization. To that purpose, we
introduced graded classification functions which encode substantial parts
of the reference relation for SOCCER on an abstract level of description,
hidden in the operations visualize and recognize (Figures 14 and 19). Can
we bring them to a more concrete level which also expresses the relationship between these two operations?
To that purpose, a new data type is introduced which explicitly represents the graded classification functions. It is called TYPOF Typicality
Potential Field (cf. Figure 20). This data type stands for the third part
of the reference relationship, the 'arch' between the two 'columns of the
bridge', namely images and propositions. As mentioned in Section 1, for
SOCCER, this part has to be essentially procedural; SOCCER does not
know (in the strict sense) anything about this data type and only can
apply the corresponding operations. But both the recognize and the
visualize operations can now be expressed simply by using TyPoFs.
494
Jrg R. J. Schirra
Figure 20. Signature of the spatial concept system (III): TyPoFs
Each proposition is associated with exactly one TyPoF (cf. operations

prop? and TyPoF? in Figure 20). The recognize operation simply can be
expressed now by asking for the value of the classification function
(Formula 3).
3.
recognize (proposition, referent)
get-T-value (referent,
TyPoF? (proposition))
As mentioned above, the visualize operation is to find the maxima of

the classification function: therefore, an operation gradient is defined. It
yields the component-wise differentiation of the classification function
for the position of the LO in the percept. This two-dimensional vector

always points in the direction of the closest local maximum. Its length is
proportional to the local slope of the classification function. At the
maximum, we get the zero vector. If the initial position of the LO is
favorable, we find the maximum of the classification function by moving
the LO in the direction of the local gradient. One step of this hill climbing
495
algorithm for one proposition is indicated in Figure 20 by the operation

approx-1: the considered object is shifted in the referent a little bit in the
direction of the given vector (cf. Yamada et al. 1988, and note 16). If we
iterate the combination of these two operations as shown in Formulas 4
to 6 until the gradient is zero, we find the maximum, or at least a local
maximum depending on the initial position of the LO.
4.
V, > 0: recognize (prop, image, + ,) > recognize (prop, image,)
5.
image 0 init-image (arguments? (prop))
6.
image^ + (prop) approx-1 (gradient (TyPoF? (prop), image,

(prop)) image, (prop), LO? (prop))
To get faster and better results, we can use typical positions associated
with the object type of LO. For example, the goal keeper typically will
be located near or in the goal, which usually is already close to the most
typical position in a particular case. Thus, the approximation of his
position should start with a position in the goal. In A N T L I M A , we
of
ApproK-Activ O b j e c t s
Cleat
Configurations
Oemo
Events
Fit
Run
Sequences
Snow
Typof Ediio'
i n (SO
^ ' ) it f w U 6 iVMlii
i n (3S 488453
4S 102077} a i t t - v a l u e 0 7784682S
l-RIGtfT-DEFEHDER i n (40 438843
42 1S8833) i t t - v a l u e 0 969099t b e e n d e t
Figure 21. Demonstration of the approximation for different starting points
496
Jrg R. J. Schirra
usually will use the positions of the objects at the time quantum before
as the initial position. Then, velocity restrictions can give further hints
as to where the objects will be (cf. note 17).
Since the proposed algorithm reacts sensitively on the starting position,
the influence of the context conditions on the localization is rather
naturally included. Figure 21 demonstrates the approximation for [near
CenterCircleLine] for several starting positions. Depending on the starting
positions which play the role of conditions of context, different solutions
of the visualization problem are constructed. Furthermore, an object to
be located with respect to a second object to be located will follow that
second LO until both have reached their optimal positions. Figure 22
shows the approximation paths for two players localized by (left of player5 player-8 (extrinsic from the lower left corner)) and (in front of player8 LeftGoal). After about 5 steps, player-5 already reached a position left
to player-8, and has to follow him until he also has reached his final
position.
The operation TyPoF-Addition combines several TyPoFs for one LO
by arithmetic average in order to simplify the approximation by reducing
Graphic Pana
Figure 22. Demonstration of the approximation for depending objects
497
the number of TyPoFs to be considered simultaneously. The basis for

this operation was described in Section 5.
Remember that TyPoFs are associated with individual propositions,
like (left player-3 player-5). Since there are quite a lot of these individual
propositions even in such a simple system as SOCCER, the task of
determining every single TyPoF would be too complicated. In fact, the
reference relation depends on all parts of the proposition: that is, both
the type of the relation, i.e., the Spatial Concept, and the arguments
influence the shape of the typicality function. Can we extract from their
instances unique descriptors of the influence of the Spatial Concepts, and
thus, separate both kinds of influences in order to get a formalism which
is easier to deal with? Although the typicality distribution for different
propositions of one type are more or less similar, the kinds of objects
and especially their dimensionality, size, and shape modify the typicality
distribution of a single proposition, stretching it, adapting it to the shape
of the RO (cf. Figures 11 to 13). Therefore, the assumed influence of the
Figure 23. Signature of the spatial concept system (IV): Typicality schemata
498
Jrg R. J. Schirra
arguments, i.e., the objects involved, has to transform the proposed

influence of the Spatial Concept underlying all of its instances.
To cover the similarity between all instances of a Spatial Concept, we
assume one function for each Spatial Concept which describes the typicality distribution not with respect to the coordinates of the objects, but
with respect to what I call "essential parameters". The typicality distribution of every instance of a spatial relation is derived from this function.
It differs from the distributions of other instances only with respect to
how the essential parameters are calculated from the objects' coordinates.
The ADT for these functions is called Typicality Schema (cf. Figure 23).
The essential parameters, e. g., distance, angles, and scaling factors, are
abstractions from the concrete coordinates. Figure 24 visualizes three
Typicality Schemata i.e., functions from essential parameters to Tvalues. For example, the Typicality Schemata for proximity and contiguity
(or the near and at concepts) both depend on two essential parameters:
the distance between LO and RO and a scaling factor (cf. note 18). How
both parameters are connected to the actual coordinates of LO and RO
depends on the dimensionality and size of the objects.
Figure 24. Three examples for typicality schemata
499
Typicality schemata can be combined by multiplication. In ANTLIMA,

the Typicality Schemata of the projective relations "being left o f ' , "
right o f ' , " behind", and " in front o f ' are defined as combinations
of the Direction and the Proximity Schemata, and thus, have five essential
parameters: distance r, distance scaling G, the reference system , the
angle 0 with respect to , and the scaling of the angle G'. In summary,
VITRAs Spatial Concepts are essentially defined as combinations of
simple graded functions (multiplication of basic Typicality schemata) of
essential parameters (cf. note 19).
The influence of the objects is encoded by TyPoF Instantiation Rules
(for short: I-rules). In addition to its Typicality Schema, each spatial
concept is associated with a set of I-rules. We can say that, in a way, Irules spread the typicality distributions encoded in the Typicality Schema
around the ROs in the percept, thus developing the appropriate Typicality
Potential Field, which, then, directs the LO towards its optimal position.
Each I-rule specifies, according to the object attributes, a set of functions
for calculating the essential parameters. These "parameter functions" as
I call them, transform the coordinates of the objects, i. e., the information
in the percept, to the essential parameters needed by the Typicality
Schemata. Since, for example, Distance, the essential parameter of the
Proximity Schema, is defined only between zero-dimensional objects, we
have to reduce or idealize higher-dimensional objects to points in
order to apply the Proximity Schema. In other words, to find the distance
between two two-dimensional objects, we first have to look for those two
points of the borders of both objects which are closest to each other (cf.
Figure 25, and also Figures 11 to 13). These points are viewed as those
objects, their distance is the distance between the objects. Actually, this
object coercion is the important part of the parameter function, since the
calculation of distance per se remains unchanged in every I-rule: the
Euclidean distance between two points.
Similarly, Angle in the Direction Schema is only defined for two lines
the zero line of the reference system and the line between RO and
LO. The geometric situations in the percept have to be adapted to this
restriction: e. g., the zero direction of the reference system can be derived
either from inherent object properties of the RO (intrinsic use) or from
another object in the context (extrinsic use) which defines exactly one
line to the RO (cf. note 20). Again, the coercion of the original percept
to an idealization plays the major part of the parameter functions used
for the Direction Schema. In every particular case, the angle is calculated
in the same way from the idealization.
500
J r g R. J. Schirra
rtiffemnl
ideal cafi
Scaling Factors
points
dualizing the
general caw;
object
Figure 25. Illustration for distance f u n c t i o n s and scaling f a c t o r s
For both cases, I-rules have to be defined. The size of the objects,
which was ignored by the essential parameters Distance and Angle, has
an influence by means of the scaling factors. For [near player-5] and
[near PenaltyArea], the Typicality Schema of Proximity not only has been
spread differently around the ROs with respect to their shape, but also
'stretched' to different diameters corresponding to the size of the ROs
(cf. Figure 25, and also Figures 11 to 13).
In summary, hidden in the I-rules of a spatial concept in VITRA are
functions which coerce parts of the percept to a more abstract, sketchlike form which is the basis for calculating the essential parameters for
the Typicality Schemata. This sketch is not explicitly modeled in VITRA
yet, though, and further studies have to be carried out about their relation
to reference semantics of spatial concepts (cf. Section 1).
The whole algorithm for visualizing elementary static spatial relations
can be sketched as in Figure 26. Originally starting from the temporally
ordered sets of elementary spatio-temporal propositions (the Propositional Elementary Structure), we consider here only one time quantum
and the corresponding set of static spatial propositions. For each proposition in this set, we find the type and the associated Typicality Schema
(operation: TySc? (type? (p))); additionally, we get the set of I-rules of
the Spatial Concept (operation: I-rules (type? (p))). The aruments of the
proposition choose one of the I-rules, which, then, transforms the Typi-
501
Propositions Elementary Structure
propos-
propos-2
TyPoF-
Spatial
Concept
Typicality
Schema
Initial Mental Image
transforms
chooses
TyPof-2
Final Mental Image

Referent for t-6
Figure 26. Algorithm of A N T L I M A ' s visualization of static spatial propositions
cality Schema to the appropriate TyPoF by specifying the appropriate

parameter functions:
7.
TyPoF? () apply-I-rule (choose (I-rules? (type? (p)),

Arguments? (p)),
TySc? (type? (p)))
By TyPoF-addition, all TyPoFs belonging to one LO in the set are

combined. Thus, we finally consider a set of LOs, each associated with
one TyPoF. With the init-image operation, we then construct a first mental
image with all the objects used as arguments in the original set of
propositions (cf. note 21). The aforementioned hill climbing approximation (cf. Formula 6) transforms this initial image to the most typical
502
Jrg R. J. Schirra
image, the referent we looked for. T h e n , the sum of all T-values finally
reached is used as a first a p p r o x i m a t i o n of the plausibility of the description f o r the listeners.
7. A comparison with Herskovits' analysis of the semantics

of spatial prepositions
O n first view, the separation of the influences on the typicality distributions described in the previous section seems to reflect the distinction
between several aspects A. Herskovits introduced in her analysis of the
semantics of spatial prepositions in English. Is this similarity only superficial, or can we also interpret the operational f o r m of the reference
relationship in S O C C E R as a simplified realization of Herskovits' analysis? T h e following c o m p a r i s o n is still on a very coarse level a n d will be
elaborated f u r t h e r in the future.
In order to avoid polysemy f o r spatial prepositions, Herskovits (1986:
39) suggests an ideal meaning of a spatial preposition underlying all uses
of t h a t preposition. In her own words:
The ideal meaning of a preposition is a geometrical idea, from which all
uses of that preposition derive by means of various adaptations and shifts.
An ideal meaning is generally a relation between two or three ideal geometric objects (e.g., points, lines, surfaces, volumes, vectors) in fact,
ideal meanings are usually those simple relations that most linguists and
workers in artificial intelligence have proposed as meanings of the prepositions. These relations play indeed an important role, but as something
akin to prototypes, not as truth-conditional meanings.
Coincidence of points, inclusion of a point in a line or in an area,
contiguity of two surfaces these are some examples of relations used
as ideal meanings. A substantial p o i n t is m a d e in the last sentence of the
quote: a l t h o u g h the ideal m e a n i n g underlies every occurrence of a spatial
preposition, it m a y be " d e f o r m e d " , " w e a k e n e d " , partially "overwritten"
or " e x t e n d e d " a n d serves merely as a "gravitational center" f o r the
meanings of different occurrences, n o t as an accurate and complete
description. Actually, for Herskovits, the meaning of single occurrences
of a spatial preposition is derived f r o m the ideal m e a n i n g by three steps
(cf. F o r m u l a 8):
503
In a particular use of a preposition, the ideal meaning may have been

transferred to another relation, one that is in some way closely related;
[step 1]
this new relation may in turn be only approximately true.
[step 2]
Moreover, the objects related are mapped onto geometric objects (matching
the categories specified for the arguments of the ideal meanings) by processes of geometric imagination, idealization and selection.
[step 3]
These mappings onto geometric descriptions, corresponding to various
geometric conceptualizations and metonymies, are accomplished by a variety of functions. [Herskovits 1986: 40, with comments by JS]
A n d later, defining technical terms f o r the t r a n s f o r m a t i o n s :
The geometric meaning of a locative expression is thus a proposition
involving a relation applying to geometric descriptions [3] of the objects,
and that relation may be the result of two transformations applying in
succession to the ideal meaning, which I call sense shifts [1] and tolerance
shifts [2], (Herskovits 1986: 40, numbers added by JS)
8.
[T [S [IM]]] (G, (Of), G 2 (02))
with:
S
G,
IM
Tolerance Shift (step 2)

Sense Shift - (step 1)
G e o m e t r i c Idealization of Object O, (step 3)
Ideal M e a n i n g
Although the spatial prepositions in S O C C E R ' s language are quite

restricted c o m p a r e d to o u r everyday language use, the reference-semantical mechanism controlling this reduced language is already r a t h e r c o m plicated, as we saw. It is m y thesis that this m e c h a n i s m can be m a t c h e d
with parts of Herskovits' analysis.
Already on first view, o n e notices a relationship between Spatial C o n cepts in S O C C E R a n d ideal meanings. Spatial concepts represent the
meaning of spatial prepositions per se, so to speak; they f u n c t i o n as
reference points with respect to which p r o p o s i t i o n s a b o u t the perceived
spatial e n v i r o n m e n t are f o r m u l a t e d . In general, they are defined by
c o m b i n a t i o n s of Typicality Schemata. Similarly, an ideal m e a n i n g stands
f o r the m e a n i n g of a spatial preposition itself, a n d is defined as a
conjunction of elementary spatial relations which are "perceptually salient
relations", i. e., "easily, quickly a n d accurately perceivable" (Herskovits
1986: 54, a n d n o t e 22). O f course, it is n o t just by chance t h a t m y
examples f o r Typicality S c h e m a t a Proximity, Contiguity all can be
f o u n d in the list of examples f o r elementary relations used as ideal
504
Jrg R. J. Schirra
meanings. Indeed, also the procedural aspect of ideal meanings is stressed:

"We can conceive of this core schema [i. e., the ideal meaning of 'in' (JS)]
as a logical predicate, but its psychological realization is most certainly
a routine [Ullman 1985], a procedure which checks whether spatial inclusion holds" (Herskovits 1989: 13).
Though I could not find an explicit record, it seems probable to me
that Herskovits' elementary relations correspond to mathematical concepts which encode clear, binary classifications: a set of arguments either
stands in a given relation, or it does not (cf. note 23). The second step
of adaptation mentioned above, the tolerance shift, supports this assumption. We have already considered the fuzziness of the applicability
of spatial prepositions in Section 4. To cover this phenomenon of ordinary
language with the mathematically accurate definitions of ideal meanings,
the sharp borders of applicability of the mathematical relations have to
be softened. Then, coincidence of points can be true in a weaker sense
or in other words: the relation is less applicable for two points
which do not actually coincide but only are close together.
In contrast to that, SOCCER already considers the fuzziness on the
level of Spatial Concepts: Typicality Schemata describe/define weak, fuzzy
forms of some of the mathematical relations mentioned above. This may
reflect a more empiricist view, assuming that the spatial concepts somehow evolve from perception, which is assumed to be always vague or
fuzzy. There is no obvious reason why spatial concepts should lose this
quality. Herskovits, on the other hand, seems to adopt a more rationalist
point of view, starting by a clear, simple, and especially binary classification which is somehow innate, and which afterwards has to be blurred
in order to fit the phenomena (cf. note 24). As a consequence, there is
no need for a distinction of the second type of transformation Herskovits
mentioned, the tolerance shifts, from other kinds of transformations in
SOCCER. Every transformation of Typicality Schemata necessarily
changes the "tolerance" for the applicability of that preposition in a
particular manner, e. g., spreading and stretching the Typicality Schema
around the RO.
In principle, something equivalent to sense shifts, the first step in
adapting ideal meanings to a particular instance, can occur on combinations of Typicality Schemata, as well. Skipping one of the conjuncts
or adding one more are the most prominent examples for sense shifts (cf.
Herskovits 1986: 94), and both can be performed similarly on Typicality
Schemata. Although, the domain of SOCCER is still too simple (or
505
better: SOCCER perceives it in a too improverished manner) for an

example of such a shift to be found yet.
Most interesting for this comparison is the remaining step in Herskovits' derivation scheme (no. 3), called geometric description. As mentioned
above, the objects used as arguments of the prepositions have to be
transformed to the simple geometric objects used as arguments of ideal
meanings. Since the relations defining the ideal meanings or their derivates
by sense and tolerance shifts only apply to simple geometric objects,
coercions have to be found which also can be viewed as different conceptualizations of the objects: when an object is described as 'being at'
some RO, both objects are conceptualized as points. To this purpose, a
broad set of geometric description functions was collected by Herskovits
(cf. Herskovits 1986, Chapter 5). Usually, several coercions apply in
sequence until the appropriate conceptualization is found. Pragmatic
factors like salience, relevance, and tolerance influence the decision as to
which sequence of coercions is applied (cf. also Hays 1987).
Obviously, something similar is hidden in the I-rules in VITRA; as
discussed in the previous section, I-rules specify parameter functions
which calculate the essential parameters from the percept the coordinates of the objects. To that purpose, the objects first have to be transformed to idealized forms appropriate to calculate the essential parameters points, lines etc.: a sketch-like representation of parts of the
referent is generated implicitly, so to speak. However, there is one main
difference between Herskovits and VITRA: the coercion in VITRA always
depends on all objects simultaneously, whereas in Herskovits' scheme the
geometric description functions apply to each object separately. I assume
that this difference is a consequence of the selection principles. Whereas
Herskovits considers in her much broader framework pragmatic principles which coordinate the idealizations of all involved objects, in VITRA
until now only idiosyncratic rules are defined: the condition part of Irules refers to attributes of the involved objects like dimensionality, size,
type, etc., and specifies explicitly for each relevant class of combinations
one set of parameter functions with the corresponding coercions. General
rules have not been considered yet (cf. Formulas 9 and 10).
9.
TS [PF, (LO, RO), PF 2 (LO, RO), ... PF (LO, RO)]
10.
PF, (LO, RO) = F (Idealize,, (LO), Idealize, 2 (RO))
with:
TS Typicality Schema
506
Jrg R. J. Schirra
PF, Parameter Function of the i'h Essential Parameter

F the pure parameter function (e.g., distance of points)
To conclude, the solution of the visualization problem in VITRA has
led us to a conception of the reference semantics of spatial prepositions
which is comparable to a simplified form of A. Herskovits' theoretical
framework (cf. Formulas 8, 9 and 10). As in her approach, we assume a
core meaning of each spatial preposition the Spatial Concept with its
defining Typicality Schema. Individual occurrences are derived by spreading the schema appropriately over the mental image by means of I-rules.
This spreading can also be viewed as idealizing the objects in the mental
image to simple geometric objects which corresponds closely to Herskovits' derivation step 3. Because VITRA alludes to a totally operational
form of reference semantics for spatial prepositions, the comparison offers
a computational realization of a substantial part of Herskovits' analysis.
Furthermore, using this realization to derive additional hints for future
elaborations of the framework seems a plausible path to follow.
8. Summary
The problem of integrating vision and natural language systems has led
us in the project VITRA to investigate the nature of the reference
relationship (cf. Figure 27). This resulted in a rejection of the objectivist
view due to its rather obvious weak points: essentially, objectivist reference theories cannot explain adequately how the objects in the world,
Figure 27. The reference relationship final view
507
which are viewed as the referents, conduct their influence on verbal

behavior. Instead, we view mental entities like percepts and mental images
as referents, thus adopting in VITRA an experientialist theory of the
reference relationship (cf. note 25).
In the system SOCCER, this decision resulted in an operational realization of the reference relation proper: algorithms are being developed
that bind exactly in the spirit of experiential reference semantics the
usages of spatial prepositions to SOCCER's percepts, the results of an
image understanding system. The core of these algorithms is always a
classification function which associates the percepts to a graded Spatial
Concept. A degree of applicability allows us to distinguish between good
and less good occurrences of such a graded concept and is reflected on
the verbal surface by linguistic hedges.
Combining the demands of experiential reference semantics with the
Gricean Cooperative Principle furthermore compelled us to reconstruct
in the form of mental images the referents of the text SOCCER plans to
generate in order to anticipate the understanding of the listeners. SOCCER, thus, is able to improve its texts. This leads directly to the visualization problem: since we usually lose information when we transform
percepts to propositions, the inverse transformation from propositions
to mental images has to produce additional information somehow. In
fact, both transformations have to use the reference relation: reinterpreting it, i. e., the classification functions, as typicality functions and correspondingly the degrees of applicability as degrees of typicality allows us
to solve this task by generating the most typical mental images as
reconstructed referents.
The operational realization of the reference relation in VITRA rests
on the classification/typicality functions: Typicality Potential Fields encode these functions, and are used both for recognizing spatial relations
from percepts and for reconstructing mental images from propositional
descriptions. Furthermore, we distinguish between two components which
define the TyPoF for a particular case: Typicality Schemata bring in the
general influence of the spatial concept and are transformed by I-Rules
that depend on the objects involved. Although technical reasons originally
forced this distinction, the separation of the two components seems to
reflect a deeper, cognitively important structure of the reference relation;
a comparison with the analysis of A. Herskovits revealed a significant
degree of similarity between ideal meanings and Typicality Schemata on
the one hand, and geometric descriptions of objects and parts of the
transformations in the I-rules on the other. Sketchlike idealizations which
508
Jrg R. J. Schirra
are hidden in the I-rules in VITRA and rather overtly included in

Herskovits' framework point to structures integral to the reference relationship which are connected to visual abstraction and metaphoric extension; these will have to be examined further.
In summary, with its system SOCCER, VITRA demonstrates albeit
still on a very coarse and primitive level the benefits of investigating
reference phenomena in an operational form.
Acknowledgements
I would like to thank heartily Ellen Hays who not only had a critical look at my English
but "obstetrically" helped me to clarify my ideas during hours of discussion. Also my
colleagues in Saarbrcken are to be thanked for comments and critiques on earlier versions.
Furthermore, all participants of the workshop on the Semantics of Prepositions in Saarbrcken 1990 contributed with their discussion to the present form of the paper.
Notes
1. General arguments against every realistic approach of reference can be found in
Empiricus (1985) and Wittgenstein (1963); both authors base their arguments on a
reflexive turn: asking how we could refer to the reference relationship itself they are
led to sceptical consequences which "destroy" the presuppositions of objectivist reference theories, namely the possibility of access to an independently given world.
2. I use the expression "mental image" not only to refer to visual images although only
those are considered in this paper; since they are connected so closely to perception,
mental images seemingly can arise in all modalities of perception; cf. Lakoff (1987:
444): "The term 'image' is not intended here to be limited to visual images. We also
have auditory images, olfactory images, and images of how forces act upon us."
3. By the way: "propositions" in this context might best be related to "symbolic cognitions"
in the sense of Leibniz (cf. Leibniz 1937, 24); in connection with corresponding
referents, they become "adequate cognitions"; the expression "reference relation proper"
corresponds closely to the term "concept" in the traditional terminology; cf. Schirra
(1993, Sect. 1.4).
4. This usage is closely connected to the procedural/declarative distinction; cf. Winograd
(1975); cf. also the distinction of different uses of the expression "proposition" in
Johnson (1987: 3).
5. This does not exclude (mental) entities to be used at one time as referents and at
another time as propositions. This is typically demonstrated by (percepts of) sketches,
e. g., of the sun: the sketch itself is a concrete thing which can be perceived and used
as referent as for example in this sentence. But normally, we do not look at the
sketch itself, but at what it represents. We use the sketch as a set of propositions about
something else the referent of the sketch which is the sun in our example (cf. Eco
1973: 208). Similarly, when speaking about percepts the corresponding mental entities
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
509
are used as propositions: the percepts of something. In this article, no such metamorphosis is considered.
More details of the system REPLAI-II which also includes strategies to recognize plan
interaction and failure are to be found in Retz-Schmidt (1991 a).
Additionally, but not described in this report, the listener model is used to control the
generation of noun phrases, anaphora, and ellipses; cf. Jameson and Wahlster (1982);
Andre (1988); and Schirra (1991).
This does not mean that listeners of broadcasted sports reports always generate visual
mental images. The reconstruction of the referents constitutes an already very high
level of understanding; cf. Craik and Lockheart (1972); and also Leibniz (1937 24).
Though, these listeners are assumed to function mentally in a way corresponding to
the simplicity of SOCCER; i. e., they perceive similarly to S O C C E R and know more
or less the same about spatio-temporal relations and composed events. If radio reporters
could not rely on such an assumption of cognitive equivalence with their listeners, sports
reports would have to be of a very different form than they actually are.
"LO" and " R O " are the most common abbreviations used, e.g., in Habel and Pribbenow (1988) and Herskovits (1986); alternative terms used in linguistics are listed in
Retz-Schmidt (1988).
This influence is documented for example in: Andre et al. (1985, 1986 a, 1986 b, 1987,
1989); Wahlster 1988; Schirra et al. 1987; Retz-Schmidt 1988; Schirra 1990 a, 1989;
Hays 1990, 1989; and Herzog et al. 1990.
The input for the complete visualization task is the temporally ordered sequence of
those sets (plus elementary velocity restrictions), called the propositional elementary
structure; the complete referent of a composed event proposition is constructed by
chunking static images appropriately together to form an image sequence.
In the right window, a sectional view cutting through the typicality cloud at its maximum
is shown, called "Traverse" and following the arrow in the graphic pane from the right
to the left; the maximum for this localizing combination is reached at a T-value of
approximately 0.9.
Dotted arrows mark those operations described earlier.
This is not completely true, but it will suffice here; cf. Section 7; In S O C C E R , all
mobile objects are zero-dimensional. Thus, it is easy to define the position of such an
object. An object of higher dimensionality could be idealized to its "center of gravity"
or a particular point of its border (e.g., the one closest to the RO) which represents
the object's location. The extension of such objects is considered by different mechanisms (cf. I-rules in Section 6).
16. The term typicality potential field is motivated by interpreting the classification function
physically as a potential field whose associated force field (the vector field resulting the
gradient operation) pulls the LO into its optimal position (with maximal potential
energy); similar to a negative field of gravity, the TyPoFs pull the LOs always "uphill
to the summit of the typicality mountains". Note that the moving of the imaginative
LO during approximation also can be used to direct (visually) searching for the
perceived LO: if we want to find an object which position is described verbally, our
visual focus of attention moves as if controlled by the typicality distribution
toward the most typical positions associated with the verbal description.
17. These hints for finding good starting positions have to be used in the aforementioned
init-image operation; cf. Schirra (1993).
510
Jrg R. J. Schirra
18. Thus, the Typicality Schemata are actually two-dimensional; the scaling dimension was
skipped in the sketch as an autonomous axis for the sake of simplicity.
19. Obviously, the combination of TyPoFs (which is actually a renormalized addition of
functions) and the combination of Typicality Schemata (which is a plain multiplication
of functions) serve different purposes: the first combination integrates the simultaneous
influences of several restrictions for one LO, whereas the second combination allows
for defining compound Spatial Concepts.
20. A special case of extrinsic use is the deictic use: the object that determines the orientation
is the speaker/listener and is normally not explicitly mentioned; cf. Retz-Schmidt (1988).
21. If a whole image sequence is to be constructed, this step is only needed at the beginning
or when a new object enters the scene; otherwise, the image for one time quantum is
used as the initial image for the approximation of the next time quantum; with this,
the algorithm always automatically consideres the context.
22. In Herskovits (1986: 55), one finds a list of relations Herskovits takes into consideration:
"enclosure, contiguity with line or surface, order of three points on a line, order of
two points on an oriented line, coincidence of two points, line in/on a plane, alignment
of points, parallelism of lines, alignment with direction, orthogonality of lines, support,
on line of sight, on orthogonal to line of sight".
23. Remember that these relations apply to geometric concepts derived from the argument
objects; Herskovits' discussion of "at" (Herskovits 1986: 51) and "to the right" (p. 184)
as "graded concepts" points toward an understanding similar to VITRA's; although it
remains unclear whether the gradation is part of the ideal meaning or not. At least,
the ideal meaning of "at" (p. 128) is simply defined as: "for a point to coincide with
another". Similarly, the ideal meanings finally proposed for the projective prepositions
(p. 190) do not include gradation: the LO has to be located (exactly) on the corresponding axis of the reference system.
24. This presentation of Herskovits' handling of tolerance phenomena is rather simplified;
cf. especially Herskovits (1986, Chapter 6.3, 'Tolerance and Idealization'); the "blurring"
is assumed to be either idiosyncratic or controlled by pragmatic factors.
25. Further investigations considering some problems inherent to the mentalist point of
view of the reference relation are to be found in Schirra (1992 a); cf. also Tugendhat
(1982, Section 20) and Schirra (1992 b).
(Figures 1 and 2 are produced on the base of two drawings of Charles M. Schultz and a
picture of the earth taken from LIFE and Banett)
References
Andre, Elisabeth
1988
Generierung natrlichsprachlicher uerungen zur simultanen Beschreibung
von zeitvernderlichen Szenen: Das System SOCCER. [Memo 28, Fachbereich Informatik, Universitt des Saarlandes, Saarbrcken, August 1988.
Diplomarbeit.]
Andre, Elisabeth Guido Bosch Gerd Herzog Thomas Rist
1985
CITYTOUR Ein natrlichsprachliches Anfragesystem zur Evaluierung
rumlicher Prpositionen. Abschlubericht des Fortgeschrittenenpraktikums,
Fachbereich Informatik, Universitt des Saarlandes, Saarbrcken.

1986 a
511
"Characterizing trajectories of moving objects using natural language path

descriptions", in: Proceedings of the 7th European Conference on Artificial
Intelligence, Brighton, U K , 1-8.
1986 b
"Coping with the intrinsic and deictic use of spatial prepositions", in: P.
Jorrand - V. Sgurey (eds.); AIMSA-86, 375-382. Amsterdam: North-Holland.
Andre, Elisabeth Thomas Rist Gerd Herzog
1987
"Generierung natrlichsprachlicher uerungen zur simultanen Beschreibung
von zeitvernderlichen Szenen", in: GWAI-87, 330-338. Berlin: Springer.
Andre, Elisabeth Gerd Herzog Thomas Rist
1989
Natural language access to visual data: Dealing with space and movement.
Bericht 63, SFB 314, VITRA, Universitt des Saarlandes, Saarbrcken.
Bateman, John
1990
"Finding translation equivalents: An application of grammatical metaphor",
in: Proceedings of the 13th International Conference on Computational Linguistics, Helsinki. Volume 2. 13-18.
Bierwisch, Manfred
1987
"On the grammar of local prepositions", in: M. Bierwisch W. Mosch
I. Zimmermann (eds.), Syntax, Semantik und Lexikon, 1-65. AkademieVerlag: Berlin.
Carnap, Rudolf
1933
"ber Protokollstze", Erkenntnis 3: 215-228.
Craik, F. J. M. - R. S. Lockheart
1972
"Levels of processing: A framework for memory research", Journal of Verbal
Learning and Verbal Behavior 11: 671-684.
Dankert, H.
1969
Sportsprache und Kommunikation Untersuchungen zur Struktur der Fuballsprache und zum Stil der Sportberichterstattung. Tbinger Vereinigung fr
Volkskunde e. V., Tbingen.
Dowty, David Robert Wall Stanley Peters
1981
Introduction to Montague semantics. Dordrecht: D. Reidel.
Eco, Umberto
1972
Einfhrung in die Semiotik. Mnchen: Wilhelm Fink Verlag.
Empiricus, Sextus
1985
Grundri der phyrronischen Skepsis. Frankfurt/M.: Suhrkamp.
Fillmore, Charles
1971
Toward a theory of deixis. [Technical report, Univ. of Hawaii. Paper presented
at the Pacific Conference on Contrastive Linguistics and Language Universale.]
Grabski, Michael
1990
Transfer statements as conditional constraints. [EUROTRA-D Working Paper
18, iai, Saarbrcken.]
Grice, Herbert Paul
1974
"Logic and conversation", in: P. Cole J. L. Morgan (eds.), Syntax and
semantics. Volume 3, 41-58. New York: Academic Press.
Habel, Christopher Simone Pribbenow
1988
Gebietskonstituierende Prozesse. [LILOG-Report 18, LILOG, IBM Deutschland, Stuttgart, Februar 1988.]
512
Jrg R. J. Schirra
Habel, Christopher Michael Herweg Klaus Rehkmper

1989
Raumkonzepte in Verstehensprozessen Interdisziplinre Beitrge zu Sprache
und Raum. Tbingen: Niemeyer.
Hays, Ellen M.
1987
A computational treatment of locative relations in natural language. [Technical Report MS-CIS-87-31, Department of Computer and Information Science, University of Pennsylvania, Philadelphia, PA.]
1989
"Two views of motion: On representing move events in a language-vision
system", in: D. Metzing (ed.), GWAI-89, 312-317. Berlin: Springer.
1990
"On defining motion verbs and spatial prepositions", in: C. Freksa C.
Habel (eds.), Reprsentation und Verarbeitung rumlichen Wissens, 192-206.
Berlin: Springer.
Herskovits, Annette
1985
"Semantics and pragmatics of locative expressions", Cognitive Science 9: 341
378.
1986
Language and cognition an interdisciplinary study of the prepositions in
English. Cambridge: Cambridge University.
1989
The linguistic expression of spatial knowledge. L. A. U. D. Report A 248,
Linguistic Agency University of Duisburg, Duisburg.
Herzog, Gerd Thomas Rist
1988
Simultane Interpretation und natrlichsprachliche Beschreibung zeitvernderlicher Szenen: Das System SOCCER. [Memo 25, Fachbereich Informatik,
Universitt des Saarlandes, Saarbrcken. Diplomarbeit.]
Herzog, Gerd C.-K. Sung Elisabeth Andre W. Enkelmann H.-H. Nagel
Thomas Rist Wolfgang Wahlster G. Zimmermann
1989
"Incremental natural language description of dynamic imagery", in: W.
Brauer C. Freksa (eds.), Wissensbasierte Systeme, 153-162. Berlin:
Springer.
Herzog, Gerd Thomas Rist Elisabeth Andre
1990
"Sprache und Raum: natrlichsprachlicher Zugang zu visuellen Daten", in:
C. Freksa C. Habel (eds.), Reprsentation und Verarbeitung rumlichen
Wissens, 207-220. Berlin: Springer.
Jameson, Anthony Wolfgang Wahlster
1982
"User modeling in anaphora generation: Ellipses and definite descriptions",
in. Proceedings of the 5 th European Conference on Artificial Intelligence, Orsay,
France, 222-227.
Johnson, Mark
1987
The body in the mind The bodily basis of meaning, imagination, and reason.
Lakoff, George
1972
"Hedges: A Study in meaning criteria and the logic of fuzzy concepts", in:
P. M. Perantean J. N. Levi G. C. Phares (eds.), Papers from the 8th
Regional Meeting, 183-228. Chicago: Chicago Linguistics Society.
1987
Women, fire, and dangerous things What categories reveal about the mind.
Leibnitz, Gottfried L.
1937
"Discourse on metaphysics", in: Discourse on metaphysics, correspondence
with Arnauld, and monadology. Lasalle, 111.: Open Court Publishing Company.
[Transl.: George R. Montgomery.]
513
LIFE - L. Banett
n. d.
Die Welt in der wir leben. Mnchen, Zrich: Droemersche Verlagsanstalt.
Marburger, Heinz Wahlster, Wolfgang
1983
"Case role filling as a side effect of visual search", in: Proceedings of the
EACL-38, 188-195.
Marr, David
1982
Vision: A computational investigation into human representation and processing
of visual information. New York, San Francisco: Freeman and Co.
1988
Language and Perception. Cambridge: Cambridge University Press.
Mohnhaupt, Michael David Fleet
1988
"Raum-zeitliche Filter fr eine top-down Steuerung der Bewegungsanalyse",
in: W. Hoeppner (ed.), GWAI-88, 296-305. Berlin: Springer.
Moilanen, Markku
1979
Statische lokative Prpositionen im heutigen Deutsch Wahrheits- und Gebrauchsbedingungen. Tbingen: Max Niemeyer Verlag.
Nagel, . H.
1988
"From image sequences towards conceptual descriptions", Image and Vision
Computing 6: 59-74.
Neurath, Otto
1933
"Protokollstze", Erkenntnis 3: 204-214.
Pribbenow, Simone
1988
"Vertrglichkeitsprfungen fr die Verarbeitung rumlichen Wissens", in: W.
Hoeppner (ed.), GWAI-88, 226-235. Berlin: Springer.
Pylyshyn, Zenon W.
1981
"The imagery debate: Analogue media vs. tacit knowledge", Psychological
Review 88: 16-45.
1988
"Various views on spatial prepositions", AI Magazine 9: 95-105.
1991a
Die Interpretation des Verhaltens mehrerer Akteure in Szenenfolgen. [Doctoral thesis, Fachbereich Informatik, Universitt des Saarlandes, Saarbrcken.]
1991b
"Recognizing intentions, interactions, and causes of plan failures", User
Modeling and User-Adapted Interaction 1: 73-202.
Saf, Hans
1966
Der Ausdruck des Ortes in der deutschen Sprache der Gegenwart. [Doctoral
thesis, Philosophische Fakultt der Friedrich-Schiller-Universitt, Jena.]
Schirra, Jrg R. J.
1989
"Ein erster Blick auf ANTLIMA Visualisierung statischer rumlicher
Relationen", in: D. Metzing (ed.), GWAI-89, 301-311. Berlin: Springer.
1990 a
"Einige berlegungen zu Bildvorstellungen in kognitiven Systemen", in: C.
Freksa C. Habel (eds.), Reprsentation und Verarbeitung rumlichen Wissens, 68-82. Berlin: Springer.
1990 b
"Expansion von Ereignis-Propositionen zur Visualisierung Die Grundlagen
der begrifflichen Analyse von ANTLIMA", in: H. Marburger (ed.), GWAI90, 246-256. Berlin: Springer.
1991
Zum Nutzen antizipierter Bildvorstellungen bei der sprachlichen Szenenbeschreibung. VITRA-Memo 49, SFB 314, VITRA, Universitt des Saarlandes,
Saarbrcken.
514
Jrg R. J. Schirra
1992 a
Connecting visual and verbal space Concerning mental images in a

listener Model. Bericht, SFB 314, VITRA, Universitt des Saarlandes,
Saarbrcken. [Forthcoming 1993.)
1992 b
Connecting visual and verbal space Preliminary considerations concerning the concept "mental image". Bericht 90, SFB 314, VITRA, Universitt
des Saarlandes, Saarbrcken. [To appear also in: Proceedings of the 4th
European Workshop on Space, Time, Movement and Spatio-Temporal Reasoning, Universite Paul Sabatier, Toulouse, 1993.]
forthcoming Zur Verbindung von visuellem und sprachlichem Raum Eine interdisziplinr orientierte Untersuchung im Rahmen der KI am Beispiel von Bildvorstellungen in einem Hrermodell zur Bildbeschreibung. [Doctoral thesis,
Fachbereich Informatik, Universitt des Saarlandes, Saarbrcken.]
Schirra, Jrg R. J. Guido Bosch C. K. Sung G. Zimmermann
1987
"From image sequences to natural language: A first step towards automatic
perception and description of motions", Applied AI 1: 287-305.
Schirra, Jrg R. J. Ellen M. Hays
forthcoming Solvitur ambulando An imaginative walk toward an understanding of
spatial metaphor. Bericht, SFB 314, VITRA, Universitt des Saarlandes,
Saarbrcken.
Sondheimer, Norm K.
1978
"A semantic analysis of reference to spatial entities", Linguistics and Philosophy 2: 235-280.
Sung, C.-K.
1988
"Extraktion von typischen und komplexen Vorgngen aus einer langen
Bildfolge einer Verkehrsszene", in: H. Bunke O. Kbler P. Stucki
(eds.), Mustererkennung 88. Berlin: Springer.
Talmy, Leonard
1983
"How language structures space", in: . Pick L. Acredolo (eds.), Spatial
orientation: Theory, research, and application, 225-282. New York: Plenum.
Tugendhat, Ernst
1982
Traditional and analytic philosophy: Lectures on the philosophy of language.
Cambridge: Cambridge University Press.
Ullman, Shimon
1985
"Visual routines", in: S. Pinker (ed.), Visual cognition, 97-159. Cambridge,
MA: MIT Press.
Vandeloise, Claude
1986
L'espace en franfais. Paris: Editions du Seuil.
Wahlster, Wolfgang
1988
"One word says more than a thousand pictures on the automatic
verbalization of the results of image sequence analysis systems", T.A.
Informations, Special Issue: Linguistique et Informatique en Republique
Federale Allemande.
Winograd, Terry
1975
"Frame representations and the declarative-procedural controversy", in:
D. G. Bobrow A. Collins (eds.), Representation and understanding, 3582. New York: Academic Press.
1963
Tractatus logico-philosophicus. Frankfurt/M.: Suhrkamp.
515
Yamada, Atushi Toyaki Nishida Shuji Doshita

1988
"Figuring out most plausible interpretations from spatial descriptions", in:
Proceedings of the 12th International Conference on Computational Linguistics,
Budapest, 764-769.
Zelinsky-Wibbelt, Cornelia
1990 a
"The semantic representation of spatial configurations: A conceptual motivation for generation in machine translation", in: Proceedings of the 13th
International Conference on Computational Linguistics, Helsinki. Volume 3,
299-303.
1990 b
"Using cognitive principles for the interpretation of spatial configurations in
machine translation", in: W. Hoeppner (ed.), Workshop Rumliche Alltagsumgebungen des Menschen RAUM-90, 209-220. Koblenz: Univ. KoblenzLandau.
Index
Prepositions are printed in italics: D = Dutch, = English, F = French, G = German

aan (D) 37
about (E) 75, 82, 95, 124, 306
above (E) 31, 106 ff 110, 125, 330
abstraction 2
in the mental lexicon 363
of essential properties 369
across (E) 31, 106, 119, 122, 332
active zone 39
adposition
spatial 20
after (E) 121, 315 f.
ago( E) 314
all across (E) 331, 334 f.
all along (E) 336
all around (E) 334
all over (E) 331, 334
all through (E) 334 f.
along (E) 19, 30, 119, 155 f., 336, 360,
368 f.
als(G)
301
ambiguity 161
lexical 11, 355, 385
an( G) 307,452
analysis
in natural language processing 1, 353
of source language 375
ANTLIMA 480
applicability degree 478, 486, 504
argument
external 223, 225 f., 229, 245
internal 238 f., 245
argument structure 101, 113 f., 118, 129,
139, 141
around () 1 9 , 3 0 , 1 5 5 , 3 3 4
artificial intelligence 357, 436, 441, 446
a s ( E ) 301
as soon as (E) 301
aspect 19, 168
at (E) 7, 19, 74, 76, 93, 108, 111, 127,
133, 305 f., 310, 329, 336, 371
attention
distribution of 373, 376
focus of 366
auf (G) 452
autonomy 30
away from (E) 329,336
axis
horizontal 15, 254, 288
identification of 407, 420
maximal 234
observer 1 5 , 2 2 6 , 2 5 4 , 2 6 0 , 2 6 5 , 2 7 2 ,
288
vertical 6, 15, 254, 371
axis-based sides 254
background 3, 9
base 8, 31
basic cognitive domains 357
because (E) 112
before (E) 102 ff., 106, 314, 316
behind (E) 31, 3 3 , 1 0 3 , 1 1 0 , 3 3 0
bei (G) 306, 448, 452, 463
below (E) 125,330
beneath (E) 330
beside (E) 122
between (E) 155,333,360
between ... and (E) 315
bevor (G) 316
bis (G) 309, 316 f., 321
Boolean algebra 401,415
bottom of object 257
intrinsic 281
boundedness 338
of reference objects 198 f., 201, 209 f.,
279
mental closure of, see Gestaltschlieung
bounding 5, 17, 190, 201, 368
by (E) 79, 119, 124, 128, 136, 140, 309,
317
by the time (E) 316
518
Index
C-selection 102, 126, 133

calculus of individuals 401
case
accusative 312
genitive 308
inherent 107, 136, 141
structural 107
case property 106, 122, 129, 139, 141
Case Theory 107
categorization
natural 205, 207
of perceptual data 339
semantic 4, 10, 13
category
discrete 3
functional 105
lexical 100 f., 122, 139, 143
natural 205
prototypical 21
schematic 21
semantic 4
vague 3
chains of meaning 18,76
classification function
and typicality 488, 491
binary 504
graded 478, 486, 493
close to (E) 125
coercion
of spatial objects 499, 505
cognition
adequate and symbolic 508
disposition 351
equivalence 509
Cognitive Grammar 3 f., 8, 12, 14, 18, 20,
363, 365
cognitive
routine 3
structure 339
coincidence situation 263 f., 278
compositionality 357
computational linguistics 18, 362
concept
abstract 5
autonomous lexical 29
geometrical 258, 435
spatial 5, 441, 484, 503
topological 258
conceptual information 221, 224 ff.,

229 f., 236, 245 f.
conceptual properties
functional 15
gestalt 15
structural 15
conceptual representation 279
conceptual structure (CS) 221, 236, 269,
272
conceptual variants 124, 126, 130
conceptualization 3 f., 8, 21, 179 f., 354
and specificity 352
metonymic 360
of spatial objects, see idealization
principles of object conceptualization
16, 180 f., 212, 214
condition
communicative 11
contextual 14, 481, 496, 510
necessary and sufficient 4, 11, 32, 179,
188, 190, 203-206
pragmatic 4, 9, 353, 365
configuration
of landmark 17,338
of trajector 338
spatial 16, 21, 39, 210, 367 f., 373
configurational information 344
conjunction, temporal 21
constraints
contextual 4
constraints on linguistic expressions
cognitive 352 f.
context
marked 431
non-verbal/extra-linguistic 270, 397
normal 432
context conditions
see condition contextual
context of discourse
see discourse context
conventionality 3 f., 10
cooccurrence restrictions 19, 88 f., 94,
107, 109
cross-linguistic relation 344, 352
dans (F) 16, 18, 41 f., 179, 183, 185,
187 f., 190, 192, 394, 401, 417, 422 ff.,
429, 431 f., 434
Index
deductions 394
default meaning 359
defaults
of semantic relations 358
defeated,
see knowledge, defeasible
depictorial representation 16, 23, 446,
467
depth assignment 261 f., 265, 276
derriere (F) 183
devant( F) 183
dimension adjectives 256
Dimension Interpretation 256 f., 262, 264,
288
dimension projection
mental process of 210
dimension reduction
mental process of 196, 210
dimensional expressions 249
dimensionality 5, 37, 39, 45, 50, 59, 62 f.,
159, 199, 213
and boundedness 50-53, 197, 199
designation of 250, 281
of objects 5 , 1 5 , 1 7 , 1 5 9 , 1 9 0 , 1 9 9 ,
201 f., 209, 213, 279, 449, 450
of related entities 37, 39, 62, 364, 368,
372
of the landmark 156, 159, 341, 373,
376
disambiguation
of word senses 96, 167, 353
discourse context 358, 360
discourse domain 21
discourse situation 8 f., 363, 370, 374 ff.,
385
disposition
spatial 374
distance adjective 256, 259
Distance Interpretation 256 ff., 262, 288
domain
abstract 19, 76, 354
basic 13
mental 73
semantic 8, 120, 123
spatial 89, 446
specific 13, 21
temporal 20, 368
519
domain of discourse
see discourse domain
down (E) 106, 155, 330
durch (G) 15,221-246,449
during (E) 119, 306 f., 310 f.
dynamism
in the lexicon 357
economy
in computational linguistics 362
in mental categorization 362
encounter situation 263, 277
encyclopedic
see meaning, encyclopedic
entlang (G) 449
environment 8
spatial 4
events
simultaneous 300
temporal ordering of 315
experience
physical 4
preconceptual 4
experiential approach 255
extension
of predicates 3
metaphorical 3, 10, 18, 20, 85, 380 ff.
metonymic 10, 18, 380
semantic, of prepositions 85
semantic 3, 13, 18, 19, 73, 362, 380,
382
spatial 5
temporal 5
family of concepts 13, 14
family resemblance relation
see family resemblance structure
family resemblance structure 10, 16 f., 29,
44, 56, 206, 362
feature
necessary and sufficient 29
semantic 1
figure 9, 31
figure-ground reversal 370
flexibility
in computational linguistics 362 f.
in the use of prepositions 371
lexical 355
520
Index
for () 11, 119, 140, 169, 170f., 312ff

317, 322, 359
foreground 3, 9
formalism 385, 394
inferential properties of 394
propositional 16
from (E) 75 f., 84, 103, 134, 332 f., 336
from ... behind (E) 103 f.
from ...on (E) 316
from ...to (E) 315,331,333
front of object
intrinsic 281
front-based region 266
function 6, 18, 21
graded 17
typical 372, 374, 376 f., 432, 434
functional properties 234 ff., 238 f.
fr/auf {G) 313, 318, 322
fuzziness 16, 60
gegen (G) 306
generalization 203
schematic 358
generation
in natural language processing 1,17,
21
of a semantic representation 358
genericity 10
geometric property 354, 367, 446
geometry
projective 401
gestalt
mental 4f., 225 f.
gestalt properties 222, 225 f., 233-239,
241
Gestalt Psychology 4, 9, 365
Gestaltschlieung
cognitive principle of 209 f.
process of 193, 197, 368 f.
gradation
see graded classification function 484,
489
grammaticalization 19 f., 100, 141 f.
granularity
of space in natural language 401
gravity 352, 366, 423
Grice's Cooperativity Principle 480
ground
in Gestalt psychology 9, 31
in metaphorization 384
head
lexical 101,107,111,113,118,121,
139 f.
semantic 101, 107
syntactic 101, 107
high{E)
256
highlighting
contextual constraints on 4, 360 f.
hinter (G) 252, 264
hoch (G) 258
Horizontal 266
idealization
of geometrical and topological properties 360, 368, 374, 376
of spatial objects 499, 503, 506
identification 285
image
mental 17
image schema 4, 89, 153, 364
immobility 273
implicatures 372, 430
in (D) 16 f., 35, 43, 56, 60
in (E) 7, 28, 64, 74, 78, 102 f., 111,
126 ff., 131, 133 f., 137, 143, 155, 308,
310, 329, 335, 360, 364, 377, 384
in (G) 310, 322
in back o / ( E ) 330
in front o / ( E ) 7, 30, 122, 330
inference 381,427,434
information processing 1, 2
economical 356
flexible 356
distribution of 1
functional 233 f., 236, 238 f.
prototypical 29
informativity 19, 169
inmitten (G) 450
innate structure 21
innerhalb (G) 311
interlingua 353
internal argument 443
Internal Localization Noun (ILN) 394
Index
interpretation
computational 15
linguistic 4
propositional 352 f.
semantic 352, 355
into{E)
7, 102, 330 f., 335
jenseits (G)
453
knowledge
about typical use 18
background 352, 359, 368
classificatory 363
conceptual 12, 362
contextual 12
defeasible 18
encyclopedic 13
flexible use of 357
indefeasible 18
lexical 12
pragmatic 18
semantic 12, 362
stereotypical 362
landmark (lm) 6, 9, 16, 19, 27, 153,
365 f., 369, 422 ff.
language performance 351, 357
later {E) 315
lattice theoretical analysis 17, 413
level
depictorial 461, 464
functional 17, 400, 410
geometrical 17, 397, 400 f.
grammatical 443
linguistic 444
of conceptual representation 12, 14 f.,
182, 265, 444 ff.
pragmatic 400
propositional 460 f., 474, 508
semantic 14, 446
lexical acquisition 353
lexical change 354
lexical entry 258
lexical variant 108, 123 f., 142
lexicalization 19, 124, 133, 143, 357
language-specific 11
lexicalization patterns 355
lexicon
semantic 13, 356 f.
521
LILOG 15,467
linguistic hedges 17
links (G) 252
listener model 480
localization 453
dynamic 15, 456 ff.
static 15
typical 16
located entity (LE)
see located object
located object (LO) 9, 15, 442 f., 448 ff.,
452 ff., 456, 459 f., 463 f., 466
location 15
random 160
vertical 86
logic 394
first-order predicate 18
non-monotonic 18, 363, 397, 432
order-sorted predicate 467
mass terms 414, 416
meaning extensions of prepositions
see extension, semantic of prepositions
meaning
conventional 355
encyclopedic 16, 327, 362
functional 16, 396, 435
generic 355
ideal 363 f., 371, 376, 385, 502
indexical 360
intrinsic 10, 353, 374
lexical 28, 118
linguistic 3 f., 13
prototypical 2, 14, 17, 206 f.
schematic 14
spatial 16, 384
mental closure of object boundaries
see Gestaltschlieung
mental completion of material boundaries
see Gestaltschlieung
mental image 4 7 3 , 4 8 1 , 4 8 9 , 4 9 3 , 5 0 1 ,
506
mental manipulation of object conceptualization
principles of 193, 196 f., 199, 202, 207215
mental model 444
522
Index
mereology 401, 415

meronomy 412, 416, 419, 421, 423 f.,
426 f., 430
metaphor 362, 380
and abstract meaning 384
established 384
frozen or dead 384
in computational linguistics 380
spatial 377
metaphorization 5, 209, 214 f.
metonymic 9
metonymy 5, 257, 362 f., 373, 380
logical 362
mind 1 , 2
mobility 5, 7
modularity 268, 435, 445
motion 7, 9, 82
nach (G) 315
nachdem (G) 316
naive physics 396, 401, 410, 436
natural language processing (NLP) 1 f.,
21 f., 353, 441
near () 118,122,125,137,330
neighbourhood region 253, 258, 260
side-based 288
network model 152
next (E) 122
object catagorization
processes of 6, 16, 181, 212, 214, 278
object category
see object concepts
object concepts 181,272,447
interaction with prepositions 177-217
spatial properties of 177-217, 188, 456
Object Schema (OS) 15, 234 f., 241, 254,
269, 278, 285
objects
spatial 476, 483, 497, 499
observer 31
observer-related sides 263 f.
of (E) 107, 109, 127, 129, 137, 140
off (E) 75, 119, 329, 332, 335
on (E) 7, 19, 28, 74, 77, 125, 128, 130 f.,
133, 136, 306, 329, 335, 364 f., 373, 376,
378 f.
on top / ( ) 122, 330
onto () 332, 335

/? (D) 32,156
orientation 7 ff., 15, 18, 272, 396, 408
actual 6 f.
canonical 6 f., 273, 281
contextually induced 6, 274 f.
deictic 411
fixed 273, 281
inherent 6, 274, 281
intrinsic 6 , 2 8 1 , 3 7 0 , 3 7 4 , 4 1 0
spatial 5 f.
orientation-point
see point of orientation
out o / ( E ) 75, 128 ff., 143, 155, 329,
330f., 335
over (E) 32, 83, 95, 107, 155, 330, 371 f.
parameter 14, 21, 221, 224f 230
configurational 338
innate 338, 348
of structure 338
path 15, 229, 231, 233, 239, 246
relational 338
semantic 285, 459 f.
past (E) 31
perception 17
perfectivity 158
period of time 79, 305,310
periods
simultaneous 303
perspective 2, 8, 352
perspectivization 6 f., 261, 263, 265, 273,
275, 281
canonical 276 f.
contextually induced 277
inherent 277
plasticity
in categorization 356, 371
plurals 413,418
point in time 79, 305, 308
point of departure 84
point of orientation 74, 76
metaphorical 77
point of view
observer's 6
speaker's 7, 9, 372 f.
points
simultaneous 303
Index
polysemy 10, 12, 32, 151, 354, 362 f., 380
position 9, 15
positional variation 281
postposition
Japanese 328
predicates
prepositional state 86
predication
function of 8
relational 8
scope of 8
preposition
accompaniment 80
area 85 ff.
case 20, 137 f., 140 f.
cause 93 ff.
circumstance 565 ff.
connection 79
contact 77, 85
deictic use of 7, 31, 165, 452, 464
directional 330-334
dynamic 80
enclosure 78
formal definition of 394
German projective 249-288
instrument 90
intransitive 173
intrinsic use of 7, 31
lexical 19
locative 334
manner 90
means 85, 90
movement 82
multiplex 19, 155
non-lexical 19
path 64, 81, 221, 223 f., 236
projective 7, 14, 17, 256
projective, extrinsic use of 14
proximity 75
separation 75, 84 ff.
simplex 19, 155
source 74, 84
spatial 27, 221 f., 471, 473, 479, 483,
502
static 35 f., 80, 85, 87
temporal 21, 295, 298
temporal ordering 315
vertical space 75
523
preposition stranding 138

prepositional variants 100
Primary Perceptual Space (PPS) 15, 261,
254 ff., 273
Principle
Lincencing 116
principle
of maximum trajector 432
of minimum landmark 433
relevance 8, 372
salience 8, 369 f.
tolerance 371
typicality 8, 370
principles
biological 8
cognitive 8, 16, 20, 182f., 208, 212,
214 f., 352, 367, 369, 385
see also mental manipulation of
object conceptualization
fixation 431
localization 448
of mentally creating boundary 211
of subsumption 209,212
pragmatic 8, 18, 21, 397, 430, 434 f.
spatial 441
processing
conceptual 459-467
of semantic representation 12,180,
182 ff., 187 f., 208, 211
profile 8, 31
projection
geometrical 408, 420
projection properties 101, 106, 111, 139,
141
Prototype theory 12
prototypicality 13, 62, 362
psycholinguistics 435
purpose 8
quantifiers
floated 136
quantity
of objects 18
radial structure 96, 484
reanalysis 130, 143
rear of object
intrinsic 281
524
Index
rear-based region 266

reasoning
analogical 355, 384
human 17
spatial 441
rechts (G) 252
redundancy
lexical 358, 363
reference object (RO) 9, 15 f., 252 f., 257,
260, 264 f., 285, 442 f., 448, 449-455,
459, 463
reference relation 471
experiential 473
objectivist 472, 508
realistic, see objectivist
reference semantics
in Artificial Intelligence 471
of spatial prepositions 471
operational 475, 493, 506
reference system
extrinsic 451
intrinsic 451
referent 471
fictitious 472
perceived 473
spatial 471
relation
antonymy 259, 260, 288
between trajector and landmark 374
common sense spatial 17
default 370
directional 7
dynamic 7
functional 17, 370
ordering 17
part-whole 17, 35 f., 55, 412, 416, 418,
421, 452, 463
positional 7
projective 7
spatial 8, 19, 21, 27, 456, 459
spatial universal 367
static 7, 80
temporal 409
topological 7
relational information 344
relevance
communicative 372
of parts of objects or situations 2, 18,

21, 352, 367, 373, 375 ff., 385, 433 f.,
450, 465
representation
lexical 18
semantic 3
right of (E) 360
round (E) 122,306
rule
lexical 20
S-selection 144
salience 8
of parts of objects or situations 2, 21,
352, 367, 375 f., 382, 385, 465
of shape 372 ff., 377
schema 13 f., 363
schema theory 362
schematic structure 5
schematization 21, 367, 370, 375 ff.
and translation 374
principles of 375
schematization type 368, 371, 382, 384
transfer of 381 f., 384 f.
search domain 16, 454 f., 459 f., 466
seit (G) 1 1 , 3 1 4 , 3 1 8 , 3 2 2 , 3 5 8
seit (dem) (G) 316
selection properties 108 f., 116, 123, 135
selection tree 22, 295, 319
selectional restrictions
of preposition 358
semantic form 221 f., 224, 227, 237, 243,
265, 269
semantic representation 14
of spatial prepositions 14, 16, 181 ff.,
188, 203 ff., 207 f., 211 f., 214
semantic structure 1, 229 f.
semantics
cognitive 27, 181, 206
common sense 362, 441
formal 3, 393 437
lexical 27 f., 206
modularized 17
naive 362, 396
procedural 400, 475, 478, 493, 503
prototype 205 f., 207, 358, 362
reference, see reference semantics
Index
semantics and pragmatics 3, 357, 435
situation 363
sense
abstract 19
durative 22
lexical default 362 f., 374
novel 18
polysemous 18
set-theory
and basic spatial relations 366
shape 5, 15, 407
sidedness 15
similarity judgements 206 f., 209
since (E) 11, 108, 125, 314, 316, 359
situation of discourse
see discourse situation
sobald (G) 301
S O C C E R 17, 475, 483
sous (F) 183, 434
space
absolute 401
Newtonian conception of 31
physical 31
relativistic 31
structure in natural language processing
401
specification 285
state 21
passive 86
stevent (state or event) 296
sur (F) 18, 183, 394, 401, 412, 4 1 9 ^ 2 2 ,
428, 434
synthesis
in machine translation 1, 353, 382
system
conceptual 13
image understanding 1 , 1 7
machine translation 21
natural language processing 373
question answering 446
semantic 13
text understanding 15, 17, 446
tegen (D) 317
Thematic Dependency Requirement ( T D R )
136
thematic roles
see theta roles
525
therefore (E) 112

theta binding 1 1 7 , 1 3 2
theta grid
113,118,129,145
theta identification 116
theta marking 114, 117
theta roles 235
Theta Theory 113
three-level a p p r o a c h
393-437
through (E) 19, 75, 77, 108, 155, 159, 331,
335, 373
throughout (E) 331, 334f.
time of discourse 21, 296
time of event 21
time of state 21
time-point
see point in time
time-span
see period of time
to (E) 1 2 7 , 1 3 4 , 1 3 7 , 1 4 0
to the left of {E) 31, 33, 330, 332, 336
to the right of (E) 3 1 , 3 3 , 3 3 0
tolerance 16
t o p of object 257
intrinsic 281
topic
topological property 17, 354, 367, 446
topology 369, 401
toward(s) (E) 119, 333
trajector (tr) 6, 9, 16, 19, 27, 153, 365 f.,
369 f., 395, 422 f.
transfer
in machine translation 353, 377
metaphorical 384
transitivity
of meronomies 427, 429, 431 f., 433 f.
translation 352 f.
h u m a n 351
machine ( M T ) 1 , 3 , 3 5 1 , 3 5 3 , 385
of prepositions 355
process of 375, 385
two-level a p p r o a c h to semantics 4, 8, 12,
14ff., 16, 181 f., 249, 254f., 267f., 272,
285, 444 f., 447
two-level model
see two-level approach
typicality 18, 21, 364, 367, 385, 431, 489
526
Index
Typicality Schema 498

and ideal meaning 503
TyPoF 493
Instantiation Rules 499, 505
ber (G) 252, 258, 449
uit (D) 43
um (G) 305 f., 449
under () 19, 31, 75 f., 83, 95, 330
underneath () 330
universality
in grammatical structure 337, 340
in semantic structure 20, 153
unter (G) 252
until/till (E) 103, 111, 121, 309, 316
up (E) 104, 155
usage patterns
difference in 352 f.
use-type 190, 192, 195, 197, 199, 200 f.,
208 ff.
utterance situation 21, 367, 372
vagueness 11, 17, 355, 362, 368, 373, 385
valency 19, 168
vehicle
verbalization
of visual information 17
Vertical 254, 258, 261, 272, 287 f.
via (E) 332, 336
viewpoint
see point of view
visualization 471, 488
VITRA 475
von ... (an)/ab(G)
316
von ... bis (G) 315
vor (G) 252, 264, 314, 452, 464
whrend (G) 301, 311, 317, 321
wenn (G) 301
when(E)
301
while (E) 301
wie (G) 301
with (E) 75, 80, 94, 127
within (E) 311
world-knowledge 19, 167, 397, 431, 462
X-bar theory
19, 100, 102, 130
zwischen (G) 450

zwischen ... und (G)
315

Zelinskywibbelt C Ed The Semantics of Prepositions From Ment

Enviado por

Dados do documento

Título original

Direitos autorais

Formatos disponíveis

Compartilhar este documento

Compartilhar ou incorporar documento

Opções de compartilhamento

Você considera este documento útil?

Este conteúdo é inapropriado?

Direitos autorais:

Formatos disponíveis

Zelinskywibbelt C Ed The Semantics of Prepositions From Ment

Enviado por

Direitos autorais:

Formatos disponíveis

The Semantics of Prepositions

Natural Language Processing

The Semantics of Prepositions

Mouton de Gruyter (formerly Mouton, The Hague)

Printed on acid-free paper which falls within the guidelines

Library of Congress Cataloging-in-Publication

The Semantics of prepositions : from mental processing to natural

Die Deutsche Bibliothek Cataloging-in-Publication

The semantics of prepositions : from mental processing to

Copyright 1993 by Walter de Gruyter & Co., D-10785 Berlin.

I wish to express my thanks to all authors who have contributed to this

1. Lexicalization patterns of prepositions

Dividing up physical and mental space into conceptual categories

On the grammar of lexical and non-lexical prepositions in English

2. Semantic categorization of prepositions and context

Semantic and conceptual aspects of the preposition durch

The meaning of German projective prepositions:

3. Contrastive implications of prepositions

On universality and variability in the semantics of spatial adpositions

Interpreting and translating prepositions:

4. Image understanding and prepositions

Computing the meaning of localization expressions involving

A contribution to reference semantics of spatial prepositions:

1. The linguistic relevance of prepositions

in general. Both MT and the interpretation and linguistic encoding of

2. New ways of analyzing prepositions

2.1. Prepositions and semantic categorization

schemata, which function as cognitive models for all higher-order mental

2.2. The expressive force of prepositions

by the surface of the earth and by gravity as well as by our normally

the speaker or by an object orientation which deviates from the norm

These interconnections represent the relative position, orientation or

3. The structure and use of the lexical category preposition

How finegrained is each dimension of meaning according to the

3.1. Categorization and context

3.1.1. Prototypes and schemata

of rules. Cognitive grammarians often emphasize that language is less

structure of the lexicon as part of the grammar, as there is a mutual

semantic representation of the preposition are assigned values by being

internal and external bounding and in terms of the dimensionality of the

Aurnague and Vieu can distinguish functionally different senses of the

meaning extension. Meaning extension proceeds from a prior meaning

maticalization originating in a marked metaphorical use of a form and

4. Contrastive implications of prepositions

ing profilable structures which correspond to the common cognitive

comparison as to the expressive distinctions by temporal conjunctions

"The semantics of gradation", in: Manfred Bierwisch Ewald Lang (eds.),

1. Lexicalization patterns of prepositions

The Dutch spatial preposition "in":

1. Setting the scene

a. De vaas Staat in de kast.

1.1. Spatial prepositions defined

a. the books on the table

On in (2 a) locates the books (argument x) with respect to the table

a. The children were drawing on the wall,

In (3 a) it is only the result of the children's drawing which is located

1.2. Classical vs. cognitive lexical semantics

The Dutch spatial preposition in

on the merits of the cognitive approach, it might be instructive, given

a. the keys in the cupboard

a. The child crossed the street in front of my car.

a. a ring around the bathtub

The Dutch spatial preposition in