Escolar Documentos
Profissional Documentos
Cultura Documentos
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
Edited by
Cornelia Zelinsky-Wibbelt
Mouton de Gruyter
Berlin New York 1993
Data
Data
Preface
Cornelia Zelinsky-Wibbelt
Contents
Introduction
Cornelia Zelinsky-Wibbelt
27
73
99
Prepositions:
patterns of polysemization and strategies of disambiguation
John R. Taylor
151
179
221
249
295
viii
Contents
351
393
441
471
Index
517
Introduction 1
Cornelia Zelinsky-
Wibbelt
Cornelia Zelinsky-Wibbelt
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.
Cornelia Zelinsky-Wibbelt
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.
Introduction
Cornelia Zelinsky-Wibbelt
Introduction
Cornelia Zelinsky-Wibbelt
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
10
Cornelia Zelinsky-Wibbelt
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.
Introduction
11
12
Cornelia Zelinsky-Wibbelt
Introduction
13
14
Cornelia Zelinsky-Wibbelt
Introduction
15
16
Cornelia Zelinsky-Wibbelt
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
18
Cornelia Zelinsky-Wibbelt
Introduction
19
20
Cornelia Zelinsky-Wibbelt
Introduction
21
22
Cornelia Zelinsky-Wibbelt
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
24
Cornelia Zelinsky-Wibbelt
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.
Cuyckens
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.
28
Hubert Cuyckens
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
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)
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)
31
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)
Also the Dutch SpP op, a number of whose uses are exemplified in (8),
is not amenable to a necessary and sufficient description.
(8)
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)
33
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
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
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.
3.1.1.
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)
37
38
Hubert Cuyckens
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)
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)
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
42
Hubert Cuyckens
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.'
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
44
Hubert Cuyckens
3.1.2.
45
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.'
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
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
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)
(28)
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)
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. de nagel in de balk
'the nail in the board'
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'
(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)
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)
52
Hubert Cuyckens
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)
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)
53
54
Hubert Cuyckens
55
56
Hubert Cuyckens
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
1 = "3DIM"
2 = "bounded"
3 = "porous"
4 = "non-porous"
5 = "vaguely bounded"
6 = "2DIM"
59
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.'
op zijn
buik
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)
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)
De appel is in de kom.
'The apple is in the bowl.'
62
Hubert Cuyckens
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)
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)
65
66
Hubert Cuyckens
Figure 14.
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.
Langacker, Ronald W.
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.
71
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.
[2]
[3]
AT
ON
IN
[6]
[4]
BY
THROUGH
[5]
[7]
ABOUT
WITH
[8]
[91
UNDER
OVER
FROM
[10]
OFF
[11]
OUT OF
[12]
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
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.
77
(d) area-
(c) state
( 0 circumstance
(g) cause
Figure 3. Radial network of extensions of at
78
Rene Dirven
(e) means
(d) area
(b) time-expanse
(c) state
(f) circumstance
(g) reason
Figure 4. Radial network of the extensions of on
79
(e) means
(d) area
(b) time-span
(a)
(b)
(c)
(d)
(f)
(g)
(h)
80
Rene Dirven
(d) area
(lawyer by
profession)
(f) circumstance
(by such bad weather)
Figure 6. Radial network of extensions of static by
(a)
(b)
(c)
(d)
(e)
(f)
(g)
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)
81
(e) means/instrument
(f) circumstance
(by accident)
(g) cause/agent
(idestroyed by fire/the enemy)
Figure 7. Radial network of extensions of dynamic by
(d) area
82
Rene Dirven
(a) movement in enclosure
(g) cause
Figure 10. Radial network of extensions of about
83
2.8. Over can denote static and dynamic situations and has similar
possibilities to about:
(a)
(b)
(d)
(g)
(g) cause
Figure 11. Radial network of extensions of over
(c) state
(f) circumstance
(g) cause
Figure 12. Radial network of extensions of under
84
Rene Dirven
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.
3.1. 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
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
89
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
91
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
94
Rene Dirven
95
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
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
Women, fire, and dangerous things: What categories reveal about the mind.
Chicago: University of Chicago Press.
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.
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)
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.
102
Gisa Rauh
a. X' X"*
b. X" - "* X'
(2')
(3)
(3')
(4)
(4')
(5)
(5')
(6)
(6')
103
a.
b.
c.
d.
It
It
It
It
was
was
was
was
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
(12)
a.
b.
c.
d.
John
John
John
John
105
106
Gisa Rauh
(16)
(17)
(18)
(19)
(20)
(21)
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.
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)
(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
(25)
a.
so J *pregnant
b. utterly incompetent
*tall
c. six feet tall
*pregnant
*
incompetent
d. six months pregnant
*tall
incompetent
(26)
(27)
(28)
110
Gisa Rauh
(29)
(30)
(32)
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 boy - he
*she
(34)
a. He stayed in London.
there.
*then.
b. Where did he stay? In London.
*When
112
(36)
Gisa Rauh
The proform thus, which may also replace [P + NP)-sequences and which
is inherently marked for "manner" as is illustrated in (39),
(39)
a.
b.
c.
d.
The same holds for the wh-form how, which represents a semantically
rather than syntactically specified proform and may replace various
syntactic categories:
(41)
113
114
Gisa Rauh
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.
115
(47)
In the examples under (48) the external role of is identified with the
referential role of the verbal head:
(48)
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)
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
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)
121
2.3.
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.
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)
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!
(59)
123
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)
(62)
(63)
125
(65)
(68)
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
(70)
(72)
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
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.
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
(82)
a.
b.
c.
d.
e.
129
(84)
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)
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)
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)
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)
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
133
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)
134
GisaRauh
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)
135
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)
(99)
136
Gisa Rauh
NP
/ \
NP
137
138
Gisa Rauh
V'
b.
V'
NP
NP
NP
139
5.
Summary
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
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.
143
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
in English. Cambridge: Cambridge University Press.
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.
Langacker, Ronald W.
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.
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
153
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)
154
John R.Taylor
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)
156
(5)
John R. Taylor
(8)
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
a.
b.
c.
d.
There's
There's
There's
There's
a.
b.
c.
d.
There's
There's
There's
There's
Between and over profile both simplex and multiplex relations. In (12),
the tr occupies a simplex point relative to the lm entity(ies):
(12)
(13)
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)
(15)
(17)
159
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
(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)
161
(27)
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)
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)
(33)
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)
163
a.
b.
c.
d.
PLACE only
next to, beside
at the front of
at the back of
in between
above
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)
(41)
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)
In English, also, the path function of the lm may be made more explicit:
(44)
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.
165
(45)
(46)
(47)
(48)
(49)
(50)
(51)
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)
167
a.
b.
c.
d.
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
(59)
169
minutes.
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)
However, the hypothesis is rendered suspect by the fact that the temporal
phrase in (67), with the verb fly, also has PP scope:
(67)
171
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
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.
Chicago: University of Chicago Press.
Lakoff, George
1987
Women, fire, and dangerous things: What categories reveal about the mind.
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.
175
2. Semantic categorization of
prepositions and context
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
Priska-Monika Hottenroth
can be used to refer to one and the same extra-linguistic situation. Wellknown examples of this phenomenon are the following constructions:
(1)
(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)
(4)
(5)
181
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
Priska-Monika Hottenroth
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
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
Priska-Monika Hottenroth
and
(8)
(10)
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)
(12)
(14)
186
Priska-Monika Hottenroth
(15)
(16)
(17)
(18)
(19)
(20)
187
Figure 5.
(22)
(23)
188
(24)
Priska-Monika Hottenroth
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.
(26)
(27)
(28)
(29)
(30)
(31)
(32)
(33)
(34)
(35)
(36)
(37)
(38)
(39)
(40)
(41)
189
190
Priska-Monika Hottenroth
(42)
(43)
191
(46)
{AT)
(48)
(49)
(51)
192
Priska-Monika Hottenroth
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)
(53)
(54)
(55)
193
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)
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
194
Priska-Monika Hottenroth
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)
(61)
(62)
would belong to this use-type in case the containers are not closed and,
similarly, also:
(63)
(64)
(65)
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)
(67)
(68)
195
(69)
(70)
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)
(73)
196
Priska-Monika Hottenroth
(76)
(77)
(78)
(79)
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)
(81)
g0#00 :
Figure 12. I'homme dans la foule
(83)
198
(84)
Priska-Monika Hottenroth
(85)
(87)
(89)
(90)
(91)
(92)
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)
(94)
(95)
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)
(97)
(98)
200
Priska-Monika Hottenroth
(100)
(101)
(102)
(103)
(104)
(105)
(106)
201
(108)
(109)
202
(112)
Priska-Monika Hottenroth
(114)
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
and:
(118)
204
Priska-Monika Hottenroth
(119)
(121)
(122)
(123)
(124)
(125)
(129)
205
(131)
as well as in:
(132)
and:
(133)
206
Priska-Monika Hottenroth
207
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
Priska-Monika Hottenroth
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)
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)
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
Priska-Monika Hottenroth
(141)
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)
(144)
211
212
Priska-Monika Hottenroth
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)
(146)
(147)
(148)
213
(151)
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
Priska-Monika Hottenroth
use tvDe 5:
selected region:
- 2-dlmenslonal
- bounded
(4)
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
le point dans I'angle
les maisons dans le volslnage
(4)
215
216
Priska-Monika Hottenroth
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
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14,
15.
16.
17.
218
Priska-Monika Hottenroth
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.
Bierwisch, Manfred Ewald Lang (eds.)
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.
Hottenroth, Priska-Monika
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.
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.
222
(1)
Ingrid Kaufmann
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.
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)
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)
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
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)
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)
The examples in (8) may serve as illustration for the great variety of
nouns that can provide a path.
(8)
226
Ingrid Kaufmann
2.
(9)
227
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)
(11)
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)
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.)
229
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
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.
(16)
231
232
(18)
Ingrid Kaufmann
(21)
(22)
233
234
Ingrid Kaufmann
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)
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
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:
237
(29)
238
Ingrid Kaufmann
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.
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. der
the
b. der
the
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
Park.
the park,
Stadt.
the city.
243
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
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)
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)
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)
246
Ingrid Kaufmann
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)
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.
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
251
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
a. ber-unter;
b. vor-hinter;
c. rechts links
a. darber-darunter, oben-unten,
oberhalb-unterhalb
b. davor-dahinter,
vorn-hinten
c. rechts-links davon, zur Rechten-Linken von, rechterhand-linkerhand
1.2.
Undisputed preliminaries
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
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).
254
Ewald Lang
2.
255
Basic assumptions:
The framework of spatial knowledge
256
Ewald Lang
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"
257
258
Ewald Lang
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)
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
O'
O"
s
3
-
Dimension
'
"
'
"
'
Distance
Distance /
projective preposition
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
267
(11)
a. /breit/ [ + N,
/schmal/[ + N,
b. /weit/
[ + N,
c. /rechts/ [ - N ,
/links/ [ - N ,
(12)
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.
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)
269
270
Ewald Lang
271
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
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.
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.
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
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
278
Ewald Lang
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.
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"
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
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
282
Ewald Lang
Object
Schemata
Spatial Objects
I
bounded
a. b, c
dimensionality
integration of
axes
No
< (a
dist
c >
"yard"
"hole"
"wall"
max acros
diam
"sidewalk"
"circle"
"ruler"
max i-vert
"banner"
"ledge"
vert across
"door"
f-vert dist
"doorway*
0
vert
"window"
0
i-vert
"picture"
< a
max
sub
"pole"
vert sub
"cylinder"
max
subver
"tower"
max
dist
"pipe"
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 dist
obs
"cupboard"
"well"
c )>
diam vert
dist i-obs
vert dist
diam sub
i-obs dist
< (a
dist vert
obs
"den"
Figure 6. Categorization of spatial objects at the conceptual level and inventory of object
schemata
IV
<a
c>
"elevator shaft"
i-vert 0
sub
"book"
0
vert
obs
"chest"
sub
"brick"
0
i-obs
"wound"
0
0
"coffin"
vert
283
284
Ewald Lang
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.
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, . . .
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
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
>
>
>
t>
-)
GB
80
>
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.
Bierwisch, Manfred Ewald Lang (eds.)
1987
Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven,
(studia
grammatica 26 + 27) Berlin: Akademie-Verlag.
289
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
Language and spatial cognition. An interdisciplinary study of the prepositions
in English. Cambridge: Cambridge University Press.
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.
Hottenroth, Priska-Monika
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
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.
Bree1
296
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
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
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
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
There is one result that we have not shown in the selection tree, to avoid
clutter:
300
Martin Durrell -
David Bree
2. Simultaneous events
Table 1 summarises the translation equivalents found between German
and English conjunctions. First, some typical data for these are given.
301
302
gelst.
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
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
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
um round, etc.
Ich komme um Weihnachten (herum)
towards midday
towards the end of the month
Tagesanbruch
seiner Geburt
dieser Gelegenheit
der Probe
at daybreak
at his birth
on this occasion
during the rehearsal
307
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
308
Martin Durrell -
David Bree
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
310
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
gesehen.
311
312
am Tag
in der Nacht
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.
313
314
315
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
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.
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.
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
na( dat)
voor( dat)
sinds
tot (dat)
tegen
van(af) ... tot
tussen ...en
320
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.
322
uses
historic
present
als
Sub stevent
is one off
in the past
-
a clause>
a clock time
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
Duration o f Sub
TOD
323
Time O f Discourse
Default
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
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.
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
329
a. Erika is at home.
b. Erika is on the ugly blue carpet.
c. Erika is in her bedroom.
(2)
330
Bruce W. Hawkins
a.
b.
c.
d.
e.
f.
g.
h.
331
2.2.2. TR configurations
Consider the syntactic structures in (10).
(10)
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)
332
Bruce W. Hawkins
a. I flew to Chicago.
b. I flew from Chicago.
c. I flew here via Chicago.
(13)
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
334
Bruce W. Hawkins
(19)
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
336
(22)
Bruce W. Hawkins
a.
b.
c.
d.
e.
John
John
John
John
John
However, not all analysts recognize that there is a fourth natural class
to be considered along with this group.
(23)
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
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:
b. SEPARATION:
(25)
TR configurations
a. NODE:
b. IMPERFECTIVE PATH:
c. INITIATIVE PATH:
d.
e.
f.
g.
h.
i.
j.
(26)
339
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.
340
Bruce W. Hawkins
341
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
344
Bruce W. Hawkins
(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.'
Mizuumi ni
oyogimasu.
lake
postp. swim
swim in the lake.'
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
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
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.
348
Bruce W. Hawkins
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.
Chicago: University of Chicago Press.
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.
Langacker, Ronald W.
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.
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
Cornelia Zelinsky-Wibbelt
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
Cornelia Zelinsky-Wibbelt
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.
Cornelia Zelinsky-Wibbelt
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
Cornelia Zelinsky-Wibbelt
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.
359
(2) Wir warten seit drei Stunden => We have been waiting for
hours
three
[ + T E M P O R A L EXTENSION]
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.
360
Cornelia Zelinsky-Wibbelt
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.
362
Cornelia Zelinsky-Wibbelt
363
364
Cornelia Zelinsky-Wibbelt
CATEGORY
CONTIGUITY
relation
Visually apparent
CONTIGUITY
j e c t e d t o TWO-DIMENSIONAL AREAS
365
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
Cornelia Zelinsky-Wibbelt
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
(7)
(70
(8)
(80
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.
368
Cornelia Zelinsky-Wibbelt
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
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
370
Cornelia Zelinsky-Wibbelt
371
(10)
372
Cornelia Zelinsky-Wibbelt
373
She walked through the meadow on which long grass was growing.
374
Cornelia Zelinsky-Wibbelt
SHAPE, + DIMENSION
375
376
Cornelia Zelinsky-Wibbelt
(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)
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
Cornelia Zelinsky-Wibbelt
380
Cornelia Zelinsky-Wibbelt
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
382
Cornelia Zelinsky-Wibbelt
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}
(14)
(15)
predicate
landmark
trajector
cat = prep
cat = np
cat = np
PLACE =
ABSTRACTION =
ABSTRACTION =
{ABSTRACT = SEMIOTIC}
{ABSTRACT = SEMIOTIC}
BOUNDEDNESS = COUNT
384
Cornelia Zelinsky-Wibbelt
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
Cornelia Zelinsky-Wibbelt
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).
Bierwisch, Manfred Ewald Lang (eds.)
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
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.
388
Cornelia Zelinsky-Wibbelt
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.
Miller, George A. Philip N. Johnson-Laird
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
Linguistic categorization: Prototypes in linguistic theory. Oxford: Clarendon
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.
Wittgenstein, Ludwig
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
Cornelia Zelinsky-Wibbelt
"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.)
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
(2) relations
394
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
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)
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)
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.
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)
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
402
403
(, )
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)).
deff({Y:^c(Y,i(-(X)))})
405
406
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.
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
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.
4.1. 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."
413
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' (, ) ) )
=
< })
=
= 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
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 ())
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
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.
420
421
def
422
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)
423
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
426
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)))
428
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).
430
431
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
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.
5.1.4. 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.
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
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.
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.
Pribbenow
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)
(3)
(4)
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)
(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.'
443
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.
445
446
Simone Pribbenow
447
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
448
Simone Pribbenow
449
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
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.
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
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)
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
3.
454
Simone Pribbenow
R R = (R -
O)
R N = R F U RRS U R
^ RLS
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)]
456
Simone Pribbenow
localization
expression
457
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.
458
Simone Pribbenow
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.
460
(12)
Simone Pribbenow
461
462
Simone Pribbenow
street (detail)
Figure 6. Depiction for example (13)
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)
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
465
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)
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
467
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.
Bierwisch, Manfred Ewald Lang (eds.)
1987
Grammatische und konzeptuelle Aspekte von Dimensionsadjektiven. Berlin:
Akademie-Verlag.
469
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.
Miller, George A. Philip N. Johnson-Laird
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.
Retz-Schmidt, Gudula
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.
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
473
474
Jrg R. J. Schirra
the
reference relation
proper
propositional level
Figure 3. The reference relationship a first view
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.
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
tieep t i g h t .
Gaehn
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
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
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
481
482
Jrg R. J. Schirra
Figure 8. Architecture of A N T L I M A
483
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
485
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
488
Jrg R. J. Schirra
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).
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 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
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, ))
494
Jrg R. J. Schirra
get-T-value (referent,
TyPoF? (proposition))
495
5.
6.
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
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
497
Figure 23. Signature of the spatial concept system (IV): Typicality schemata
498
Jrg R. J. Schirra
499
500
J r g R. J. Schirra
rtiffemnl
ideal cafi
Scaling Factors
points
dualizing the
general caw;
object
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
propos-
propos-2
TyPoF-
Spatial
Concept
Typicality
Schema
transforms
chooses
TyPof-2
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.
503
with:
S
G,
IM
504
Jrg R. J. Schirra
505
10.
with:
TS Typicality Schema
506
Jrg R. J. Schirra
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,
507
508
Jrg R. J. Schirra
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.
511
512
Jrg R. J. Schirra
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.
Miller, George A. Philip N. Johnson-Laird
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.
Retz-Schmidt, Gudula
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
515
Index
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
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
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
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
524
Index
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
526
Index
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
315