Escolar Documentos
Profissional Documentos
Cultura Documentos
Two Morphologies
or One?
Inection versus Word-formation
Andrew Spencer
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
28 ANDREW SPENCER
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 29
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
30 ANDREW SPENCER
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 31
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
32 ANDREW SPENCER
3 Lexical Representations
base derivate
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 33
The 3sg inected form is related to this base in part because its form
subsumes that of the base and in part because it shares all its
other properties with the base. That sharing of properties is represented
by _____ and it can be thought of as the result of a kind of identity default
function (or General Default Principle in Spencer 2013: 191; see below,
Section 6): if a morphological rule fails to specify an operation over a part
of a lexical representation, then assume that the relevant representation
remains unchanged.
Canonically derived lexemes involve a non-trivial change in all four
attributes, as shown in Table 2.2.
The {SubjectNominal} label is that of a derivational feature which
governs the (paradigmatic, regular) derivation of a subject nominalization
from a verb (cf. Stump 2001: 257). Each of the four attributes is distinct
from that of the base form of the verb lexeme (and, indeed, from any of
its inected forms). However, the form of PRINTER subsumes that of the
base form of print and, more importantly, the sem attribute of PRINTER
is the sem of PRINT with the added semantic predicate denoting the subject
of the predication.1 Hence, we can say that the base verb and its derived
nominal are lexically related (paradigmatically so, in fact).
The LI, however, can also be deployed to dene the lexical entry itself.
A lexical entry is a representation that is not dened (yet) in terms of any
feature content, but which abstracts away from all the possible inected
and derived forms. We can therefore represent a lexical entry as a function
from pairings of hLI, ui. Thus, for print we will have the lexical entry in
Table 2.3.
From the lexical relatedness schemas shown in Table 2.2 and Table 2.3 we
can see that (canonical) inection and (canonical) derivation are maximally
different types of lexical relatedness. However, the schemas suggest that
1
Semantic representations are given in a very much simplied form.
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
34 ANDREW SPENCER
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 35
2
I can say cut-and-pastable, but this is an exception that proves the rule, because it is only possible by virtue of the
fact that cut-and-paste has been reanalyzed (from a syntactically constructed coordination of verbs) as a composite
verb (hence the hyphens).
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
36 ANDREW SPENCER
3
Bauer et al. (2013: 312) cite the attested example a cool iPod-like handheld controller. Here the interpretation has
to be a handheld controller similar to the controller of an iPod and not the nonsensical (or at least uncool) a handheld
controller similar to an iPod.
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 37
1. Purpose build the form of a lexeme required build a new lexeme from
by a given syntactic context an existing lexeme
2. Syntactic determined by syntax not determined by syntax
determinism
3. Obligatoriness function is obligatory function is not obligatory
4. Productivity fully productive not fully productive
5. Transparency transparent not always transparent
6. Base inheritance all base features are inherited base features that are inherited
are limited
7. Exponence order after derivational exponent; before inectional exponent;
closes word need not close word
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
38 ANDREW SPENCER
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 39
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
40 ANDREW SPENCER
has the external syntax of an adjective, but retains the ability to express the
verbs internal arguments and may even retain idiosyncratic properties
such as assigning lexical case to its direct object (Haspelmath 1996;
Spencer 1999).
When we consider Brown and Hippisleys transparency criterion from
this vantage point, it seems that what looked strikingly obvious at rst is
actually far from clear. The problem is to avoid comparing unlikes. When
we say that an agent nominal is opaque to the inectional properties of the
base what we really mean is that there is a considerable distance between
certain typical inected forms of the base lexeme and certain typical
inected forms of the derived lexeme. But we have to be rather careful
with such comparisons. It isnt clear, for instance, that the form driving in,
say, the driving of fast cars is more transparently related to the 3sg form drives
than it is to the agent nominal driver, in say a driver of fast cars.
Property 7, exponence order, reects the undeniably very strong ten-
dency for inections to appear external to derivational markers. However,
there are well-known (and some less well-known) problems in establishing
afx ordering as a hard-and-fast principle. Stump (2001, 2005) draws atten-
tion to the phenomenon of head-application in inection. A simple example
of head-application is afforded by English prexed verbs such as under-
stand. In order to inect this verb properly we effectively have to ignore
the prex, because the lexeme inherits the irregular morphology of STAND:
stood~under-stood. However this is coded in a grammar, it essentially boils
down to the fact that we rst inect the stem for tense and then attach
the prex, contrary to Principle 7.
A less frequently discussed problem for a thoroughgoing application of
Principle 7 is that posed by discontinuous stems. It is not uncommon for a
language to have lexemes (typically verbs) whose inected stems are com-
plex, consisting of, say, a verbal root and a preverb. In a number of language
groups that preverb might be separated from the root by a string of afxes,
some of which, at least, are typical inections (for instance, subject agree-
ment markers). The Athabaskan group and its distant kin the Yenisseian
group (now represented by the isolate Ket, spoken in Siberia) are well-
known instances (see Rice 2000 for a detailed discussion of Athabaskan,
and Vajda 2004 for Ket).
6 Intermediate Categories
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 41
base derivate
In Table 2.6 the sem representation for hzban adds a semantic predicate
corresponding to the notion (be) in, but the LI representation remains
unchanged, showing that we are dealing with an inected form of the noun
lexeme.
Another type of relatedness which is difcult to classify as inection or
derivation is seen in argument structure alternations. Many language
groups have a family of alternations including passive (or antipassive),
applicative, causative, reexive/reciprocal alternations among others.
The terminological quandary posed by argument structure alternations
4
The SEM representation for the inessive can be read as the property of being some P such that the relation P is
in x holds of P and a house, x.
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
42 ANDREW SPENCER
base derivate
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 43
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
44 ANDREW SPENCER
The previous section has outlined some (but not most!) of the multifarious
types of lexical relatedness that can be identied, but has not yet addressed
the central problem: how can we unite the mechanism used to dene
inected forms of a lexeme and the mechanism used to dene new (and
newly inectable) lexemes?
The solution proposed by Brown and Hippisley (2012: Chapter 7) relies
on the concept of hierarchical network in Network Morphology (see
Chapter 18). That model denes lexical relatedness in terms of sets of
defaults and overrides. For inection these dene the inectional para-
digms, independently of lexeme selection. For derivation Brown and Hip-
pisley assume two hierarchies, a lexemic hierarchy and a derivational
hierarchy. The lexemic hierarchy is effectively the models lexicon. It
denes all the idiosyncratic information about each lexemes phonology,
morphology, syntax, and semantics. The derivational hierarchy consists of
lexeme-formation templates (LFTs). In essence, these are morphological
constructions, as in Booijs (2010) model of Construction Morphology
(Chapter 16), and fulll the same role as word formation rules in earlier
models. A derived lexeme such as driver therefore inherits information
from its base to drive and also from the orthogonal hierarchy which
includes the lexeme-formation template dening -er subject nominaliza-
tions. That template either adds content to the bases entry (the afx itself
and the additional semantic predicate) or overrides existing information
(e.g., the syntactic class of the derived word).
Derivational morphology of this sort largely conforms to the canonical
denition of derivation. Brown and Hippisley then provide a description of
three less canonical types of derivation. The rst is conversion, in which the
LFT overrides or modies all the base lexemes information except its
phonology. Hence, the derived lexeme inherits the form of its root directly
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 45
from the base lexeme. The second is the transposition, in which the LFT
fails to modify the semantic entry of the base, which therefore passes on its
meaning to the derived lexeme. The third deviation from canonical deriv-
ation is category-preserving derivation, as exemplied by evaluative morph-
ology such as Russian diminutives. In this case it is the syntactic
information which is inherited from the base lexeme.
Brown and Hippisleys Network Morphology treatment does not include
an explicit distinction between lexeme-preserving morphology and lexeme-
changing morphology. Thus, the three non-canonical types of relatedness
are each treated as ways of dening new lexemes and each requires, along
with canonical derivation, information-changing relations rather than just
information-specifying relations, as we nd in (canonical) inection.
Spencer (2013) proposes a similar solution to that of Brown and Hippis-
ley, in that he assumes that lexical relatedness is dened in terms of
defaults and overrides (Chapter 11). Both derivation and inection are
dened in terms of a paradigm function in the sense of Stump (2001)
(Chapter 17), but generalized so that it can dene form, syn, sem and LI
attributes, the Generalized Paradigm Function (GPF). At the same time the
form, syn, sem attributes are dened over pairings of hLexemic Index,
{feature set}i . The {feature set} for derivation is the label of the deriv-
ational relation itself. A basic uninected lexical entry is given by dening
the GPF over the pairing hLexemic Index, ui, where u is the empty feature
set. The specication of a LI for each lexeme permits a broader typology of
relatedness, as we have seen.
For standard inection, which does not introduce a semantic predicate
and does not change the syntactic category, the GPF will non-trivially
specify only the form properties. The syn, sem, LI attributes will therefore
remain unchanged by virtue of the General Default Principle. The GPF is
then effectively equivalent to the paradigm function of standard Paradigm
Function Morphology (PFM). However, for other types of intra-lexemic
relatedness, such as evaluative morphology, meaning-bearing inection,
and transpositions, non-trivial changes will be effected over the syn and/
or sem attributes by the fsyn and/or fsem components of the GPF.
With canonical derivation, however, the GPF effects non-trivial change in
all four attributes, as illustrated schematically in Table 2.2. Let AbilAdj be
the derivational feature governing -able adjective formation (as in printable).
Then, the pairing hPRINT, {AbilAdj}i will be mapped to the full lexical
representation of PRINTABLE. Assuming that this process is entirely regular
(i.e., paradigmatic) we can dene the LI for the derived word in terms of
the derivational category itself and the LI of the base lexeme. In other
words, we can say that the LI for PRINTABLE is not just some arbitrary
integer or label such as printable but is the value of a function, fli, which
maps hPRINT, {AbilAdj}i to hPRINTABLE, {AbilAdj}i. This is achieved by
dening fli (hPRINT, {AbilAdj}i) = fAbilAdj(PRINT), where fAbilAdj is a function
over LIs dened with respect to the derivational feature AbilAdj (see
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
46 ANDREW SPENCER
Spencer 2013: 179). We then invoke the Derived Lexical Entry Principle
(Spencer 2013: 200), according to which any GPF that introduces a non-trivial
but systematic (paradigmatic, hence productive (Chapter 4)) change in
LI of this sort gives rise to a representation which is equivalent to
hfAbilAdj(PRINT), ui . This is equivalent to saying that the label PRINTABLE is
the (derived) LI of the output lexeme, and that it is an as yet uninected lexical
representation, hence effectively the lexical entry for the derived lexeme.
Finally, we can observe that in well-behaved (that is, near-canonical) cases
of derivation we nd that the syn properties of the derived word are
projectable by default from the sem properties. To reect this Spencer
(2013: 201) proposes a principle under which the syn properties of the
output of the GPF, as well as most of the form properties, are replaced by
the empty set (i.e., effectively deleted). This representation then triggers the
operation of a general Default Cascade, under which the syn properties of
a paradigmatically derived lexeme are inherited by default from the sem
attribute (so that a word denoting a Thing is by default syntactically a noun,
while a word denoting an Event is by default a verb). Thus, the schematic
representation of the derivational GPF for er-nominalizations in English
shown in Table 2.2 can be replaced by Table 2.7.
In sum, we can individuate lexemes by means of an arbitrary (numerical)
index, LI. We can then deploy that index to dene the lexical representation
itself as well as the lexical relatedness functions that change the various
aspects of inection, derivation, and other types of relatedness. Where we
have derivational morphologythat is, where the lexical relation denes a
new lexeme and hence a new LIwe invoke the principle that the deriv-
ational paradigm function denes a representation hf(), ui, where is the
LI of the base lexeme. In this way we can unify the functions which govern
inection, derivation, and all intermediate types of relatedness, thus
addressing the question which opened this section.
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 47
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
48 ANDREW SPENCER
References
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002
Two Morphologies or One? 49
The View from Building 20: Essays Presented to Silvain Bromberger on his
50th Birthday, 11176. Cambridge, MA: MIT Press.
Haspelmath, Martin. 1996. Word-class-changing inection and morpho-
logical theory. Yearbook of Morphology 1995, 4366.
Helimski, Eugene. 1998. Selkup. In Daniel Abondolo (ed.), The Uralic
Languages, 5489. London: Routledge.
Kenesei, Istvn 2007. Semiwords and afxoids: The territory between word
and afx. Acta Linguistica Hungarica 54, 26393.
Los, Bettelou; Corrien Blom, Geert Booij, Marion Elenbaas, and Ans van
Kemenade. 2012. Morphosyntactic Change: A Comparative Study of Particles
and Prexes. Cambridge University Press.
Randall, Janet. 1989. Morphological Structure and Language Acquisition:
PhD dissertation, University of Massachusetts, Amherst. New York: Garland
Publishers.
Randall, Janet 2010. Linking: The Geometry of Argument Structure. Dordrecht:
Springer Verlag.
Rice, Keren D. 2000. Morpheme Order and Semantic Scope: Word Formation in the
Athabaskan Verb. Cambridge University Press.
Sims, Andrea. 2015. Inectional Defectiveness. Cambridge University Press.
Spencer, Andrew. 1995. Incorporation in Chukchi. Language 71, 43989.
Spencer, Andrew. 1999. Transpositions and argument structure. Yearbook of
Morphology 1998, 73102.
Spencer, Andrew. 2005. Towards a typology of mixed categories. In
C. Orhan Orgun and Peter Sells (eds.), Morphology and the Web of Gram-
mar: Essays in Memory of Steven G. Lapointe, 95138. Stanford University
Center for the Study of Language and Information.
Spencer, Andrew. 2013. Lexical Relatedness: A Paradigm-based Model. Oxford
University Press.
Stump, Gregory T. 1993. How peculiar is evaluative morphology? Journal of
Lingustics 29, 136.
Stump, Gregory T. 2001. Inectional Morphology: A Theory of Paradigm Struc-
ture. Cambridge University Press.
Stump, Gregory. 2005. Some criticisms of Carstairs-McCarthys conclusions.
Yearbook of Morphology 2005, 283303.
Vajda, Edward J. 2004. Ket. Languages of the World/Materials 204. Munich:
LINCOM EUROPA.
Williams, Edwin. 1981. On the notions lexically related and head of a
word, Linguistic Inquiry 12, 24574.
Wunderlich, Dieter, and Ray Fabri. 1995. Minimalist Morphology: An Approach
to Inection. Zeitschrift fr Sprachwissenschaft 14, 23694.
Zwicky, Arnold. 1985. How to describe inection. In Mary Niepokuj, Mary
Van Clay, Vassiliki Nikiforidou and Deborah Feder (eds.), Proceedings of
the Eleventh Annual Meeting of the Berkeley Linguistics Society, 37286.
Berkeley Linguistics Society.
Downloaded from https:/www.cambridge.org/core. University of Florida, on 10 Feb 2017 at 19:32:45, subject to the Cambridge Core terms of use, available at
https:/www.cambridge.org/core/terms. https://doi.org/10.1017/9781139814720.002