Paradigm Function Morphology And The Morphology Syntax Interface

Preparing to load PDF file. please wait...

0 of 0
100%
Paradigm Function Morphology And The Morphology Syntax Interface

Transcript Of Paradigm Function Morphology And The Morphology Syntax Interface

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 383 1.9.2006 5:00pm
chapter 12
....................................................................................................................................................
PA R A D I G M FUNCTION MORPHOLOGY AND THE MORPHOLOGY–
S Y N TA X INTERFACE
.....................................................................................................................................................
thomas stewart and gregory stump
12.1 A Word-Based Interface Between Morphology and Syntax
.........................................................................................................................................................................................
Syntactic representations the phrases and sentences of a language accommodate the insertion of morpholexical (¼ morphological or lexical) expressions drawn or projected from its lexicon. Morphologists disagree about the types of morpholexical expression that is inserted and about the types of node into which insertion

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 384 1.9.2006 5:00pm
384 thomas stewart and gregory stump
takes place. Some argue that stems and aYxes are inserted into separate nodes, so that the syntactic representation of a single word may involve several instances of morpholexical insertion; in such an approach, words have the status of syntactic complexes, so that a word’s interaction with rules of syntax is mediated by the constellation of nodes (‘‘morphemes’’) of which it is constituted.1 Others argue that words are syntactic atoms instead—that they are inserted into syntactic structure as wholes and that their own internal morphological structure is unavailable to syntactic manipulation; in this approach, a word’s interaction with rules of syntax is entirely determined by the unordered2 set of morphosyntactic properties associated with the node that it occupies. Scrutiny of the empirical evidence reveals that the latter, word-based, conception of the morphology–syntax interface is more compatible with the range of behaviors exhibited by natural-language morphology than the former, morpheme-based, conception of this interface.
One pertinent sort of evidence is the fact that words which are completely alike in their external syntax may diVer in their morphology. In the word-based approach, the syntactic behaviour of a word is associated with the unordered set of morphosyntactic properties situated at the node that it occupies; this approach therefore draws no connection between a word’s syntactic behaviour and the exponence3 of its properties. In the morpheme-based approach, by contrast, a word’s syntactic behaviour is directly tied to the conWguration of morphemes of which it is constituted; thus, the morpheme-based approach, unlike the word-based approach, predicts that words that are alike in their syntax should show similar exponence. This prediction is not borne out. English past-tense verb forms, for example, exhibit suYxal exponence (tossed), apophonic exponence (threw), extended exponence (sold), and null exponence (hit). In order to accommodate examples of the latter three types, proponents of the morpheme-based approach must assume that words that are alike in their syntax have similar morphology at some abstract level of representation but that this similarity is obscured by superWcial operations; for instance, one might assume that tossed, threw, sold, and hit share an abstract structure of the form [V Tns] but that in the latter three cases, this structure is obscured by the transformational fusion of the V and Tns nodes or by the insertion of a zero suYx. There is no independent syntactic motivation for the postulation of such operations, whose sole rationale would be to get the morphology to Wt the
1 Indeed, the very notion ‘‘word’’ is epiphenomenal in a morpheme-based approach of this type. 2 In this type of approach, a word’s morphosyntactic properties are linearly unordered. Nevertheless, we assume that there may be at least two sorts of hierarchical relations among morphosyntactic properties. First, a morphosyntactic feature may be set-valued, so that its values include speciWcations for other features, e.g. agr:{per:1, num:sg}. Second, we leave open the possibility that grammatical principles might be sensitive to a ranking relation over morphosyntactic properties; see e.g. Stump (2001: 238 V). 3 In the morphology of a word w possessing a morphosyntactic property (or property set) p, the exponents of p are those morphological markings in w whose presence is associated with that of p. Exponence is the relation between a property (or property set) and its exponent(s).

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 385 1.9.2006 5:00pm

paradigm function morphology 385
syntax; nor is there a shred of independent evidence even given their common morphosyntactic content that threw has an underlyingly aYxational structure like that of tossed. As this example suggests, the morpheme-based approach (unlike the word-based approach) is committed to the assumption that trees are as suitable for representing a word’s morphology as for representing phrasal syntax; on this approach, all morphology is seen as fundamentally aYxational, notwithstanding the prima facie counter-evidence of non-concatenative morphology of diverse sorts.
Another kind of evidence favouring the word-based conception of the morphology–syntax interface is the fact that words which diVer in their external syntax do so in ways which correlate with their content, not with their form. This fact follows from the assumptions of the word-based approach, according to which syntax may be sensitive to a word’s morphosyntactic properties but is in any event blind to its morphological form. In the morpheme-based approach, by contrast, the possibility is left open that words’ morphological structure might correlate with diVerences in their external syntax that are not simply predictable from diVerences in their morphosyntactic content. The morpheme-based conception of the morphology– syntax interface is therefore permissive without motivation. For instance, the external syntax of the Fula verbs in Table 12.1 is fully determined by the unordered sets of morphosyntactic properties with which they are associated. The syntax is simply blind to whether these verbs inXect preWxally or suYxally; the third-person singular, Wrst-person plural, and third-person plural forms do not function as a natural class with respect to any syntactic behaviour, nor do the complementary, suYxed forms. This is not an oddity of Fula; all human languages are like this.
A third type of evidence favouring the word-based conception of the morphology–syntax interface is the fact that languages diVer morphologically in ways which cannot be attributed to independently motivated diVerences in their syntax. Thus, as Stump (2001: 25 V) points out, the morphological expressions of tense and voice are in opposite orders in Albanian lahesha and Latin lava¯bar (both ‘I was

Table 12.1 Some relative past active forms of the Fula verb loot ‘wash’

Singular 1 lootu-mi’ 2 lootu-daa’ 3 ‘o-looti’

Plural
min-looti’
&
lootu-den’ (inclusive) lootu-don’ (exclusive) e-looti’

Source: Arnott 1970: 191

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 386 1.9.2006 5:00pm
386 thomas stewart and gregory stump
washed’), yet there is no independent justiWcation for claiming that the operators of tense and voice participate in contrasting relations of c-command in these two languages; similarly, the morphological expressions of tense and subject agreement are in opposite orders in Latin ama¯bam and Welsh Romany kama´vas (both ‘I loved’), yet there is no syntactic evidence for any diVerence in the nesting of TP and AgrP in these languages; and so on. The fact that the ordering the aYxes of word often corresponds to an assumed nesting of functional categories has sometimes been used to argue for the morpheme-based approach to the morphology–syntax interface (cf. Embick and Noyer, in this volume). But this tendency can be seen simply as the eVect of relevance on diachronic processes of morphologization (Bybee 1985: 38 V); indeed, the latter explanation is easier to reconcile with the frequent incidence of ‘‘exceptions’’ such as lava¯bar or kama´vas.
We conclude from this type of evidence that the interface between morphology and syntax is, in the terminology of Zwicky (1992: 356), a featural rather a formative interface—that the morphology and syntax of a language have only a limited shared vocabulary, which includes lexical categories and morphosyntactic properties but excludes such notions as aYx or inXectional morpheme (Stump 2001: 18 V). Accordingly, we believe that the adequacy of a morphological theory is, in part, a function of the extent to which it accommodates this conception of the morphology–syntax interface.
Here, we present an overview of Paradigm Function Morphology, a formally explicit morphological theory which presupposes a word-based interface between morphology and syntax. We begin by situating Paradigm Function Morphology within the general landscape of current morphological theories (sections 12.2 and 12.3), then proceed to a discussion of its central premises: the need to distinguish between content-paradigms and form-paradigms (sections 12.4 and 12.5), the need for both paradigm functions and realization rules in the deWnition of a language’s morphology (sections 12.6 and 12.7), and the centrality of Pa¯n. ini’s principle (section 12.8). In section 12.9, we return to and elaborate on the word-based conception of the morphology–syntax interface aVorded by PFM; we contrast this conception with the morpheme-based conception postulated by theories such as Distributed Morphology in order to highlight the signiWcant empirical and descriptive advantages of the PFM approach (section 12.10). We summarize our conclusions in section 12.11.
12.2 What is PFM?
.........................................................................................................................................................................................
Paradigm Function Morphology (PFM) is an inferential–realizational theory of inXectional morphology which takes as its central premise the assumption that paradigms are essential to the very deWnition the inXectional system of a language.

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 387 1.9.2006 5:00pm
paradigm function morphology 387
It is realizational because it presumes that a word’s inXectional markings are determined by the morphosyntactic properties which it carries; that is, it rejects the assumption, characteristic of incremental theories, that words acquire their morphosyntactic properties only as an eVect of acquiring the exponents of those properties. In addition, PFM is inferential because it presumes that word forms are deduced from more basic forms (roots and stems) by means of rules associating particular morphological operations with particular morphosyntactic properties; that is, it rejects the assumption, characteristic of lexical theories, that morphosyntactic properties are associated with inXectional markings just as lexico-semantic properties are associated with lexemes—in lexical entries or as ‘‘vocabulary items’’.4
The incremental–realizational distinction and the cross-cutting lexical– inferential distinction deWne theories of inXectional morphology of four logically possible types. As Stump (2001: 2f) shows, all four types are instantiated among current approaches to inXectional morphology: the lexical–incremental type is embodied by the theory advocated by Lieber (1992); the inferential–incremental type, by the theory of Articulated Morphology (Steele 1995); the lexical– realizational type, by Distributed Morphology (hereafter DM; Noyer 1992; Halle and Marantz 1993); and the inferential–realizational type, by the general approach of Word-and-Paradigm morphology (Matthews 1972; Zwicky 1985), A-morphous Morphology (Anderson 1992), Network Morphology (Corbett and Fraser 1993; Brown, Corbett, Fraser, Hippisley, and Timberlake 1996), as well as PFM.
12.3 Why an Inferential–Realizational
Theory?
.........................................................................................................................................................................................
Semiotically oriented theories of morphology such as Natural Morphology (Dressler, Mayerthaler, Panagl, and Wurzel 1987) emphasize the cognitive value of isomorphism between units of content and units of form in morphological structure. From this perspective, an ideal system would have one and only one distinct, phonologically invariant, morpheme paired with each possible distinct morphosyntactic property. The ideal is not achieved in natural human language, however, not even in highly agglutinative language types. Languages commonly and successfully exploit all kinds of deviations from the canonical one-to-one pairing of form with content. This by no means undercuts the Natural Morphology
4 Lexical theories of morphology may, however, maintain a distinction between stems and aYxes, e.g. by assuming that aYxes are inserted into syntactic structure later than stems (see Embick and Noyer, in this volume).

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 388 1.9.2006 5:00pm
388 thomas stewart and gregory stump
position; rather, it serves to point out that whatever the cognitive ideal, the morphological descriptive framework must be ready and able to allow for a range of morphological exponence beyond unifunctional, phonologically invariant, and consistently placed segmentable aYxes. PFM, by insisting on the principled separation of content and exponence, allows for the range of observed morphological behaviours without necessitating structural zeroes, ad hoc hierarchical conWgurations, or treating some types of exponence as more or less ‘‘normal’’ in absolute terms. Incremental and lexical theories are less well suited to structures that are not built up in a monotonic increasing fashion out of discrete, intrinsically meaningful pieces.
Therefore, theoretical approaches that take aYxation as basic and all other exponence as somehow deWcient (e.g. segmentally underspeciWed reduplicants, Xoating mutation features), are hard-pressed to accommodate such non-canonical exponence in the morphological description. Canonical inXection is compatible with a variety of theoretical approaches; it is the noncanonical phenomena that provide the basis for choosing among them.
Incremental theories are based on the sometimes tacit assumption that inXectional markings are added to words in order to allow them to acquire their full set of morphosyntactic properties; accordingly, this type of theory implies that extended exponence (the appearance of more than one marking for the same property or property set) should never arise, for the simple reason that it is never motivated by the need to augment a word’s morphosyntactic property set. Yet, extended exponence is widespread in inXectional morphology (Stump 2001: 3 V). For instance, the default plural suYx -ou` appears twice in Breton bagou`igou` ‘little boats’: contrary to the basic premise of incremental theories, the addition of the second -ou` is not motivated by the need to supplement the word’s morphosyntactic property set, nor is the stem bagou`ig- (whose morphosyntactic property set is, if anything, already fully speciWed) acceptable, in itself, as a word for ‘little boats’ in Breton (*bagigou` is likewise ungrammatical.) Under the assumptions of an inferential–realizational theory, there is no expectation that extended exponence should not arise, since there is no reason, a priori, why the morphology of a language should not contain two or more rules realizing the same property.
Because they portray inXectional markings as the source of a word’s morphosyntactic properties, incremental theories imply that every one of a word’s morphosyntactic properties should be interpretable as the contribution of a particular marking. But this, too, is an unsatisWed expectation in morphology: a word’s morphological form may underdetermine morphosyntactic content. Consider an example from Sora (Austroasiatic; India). In Sora, the second-person plural aYrmative non-past form of the verb de ‘get up’ is @deten ‘you (pl.) get up’; see Table 12.2. This form has an overt marking for tense (the nonpast suYx -te), a conjugation-class marker -n, and a default plural preWx @-; nowhere does it exhibit an overt exponent of second person. Yet, it is unmistakably the second-person

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 389 1.9.2006 5:00pm
paradigm function morphology 389
plural form: its Wrst-person plural inclusive counterpart is detenbe (in which the appearance of the Wrst-person plural inclusive suYx -be overrides that of @-); its Wrst-person plural exclusive counterpart is @detenay (which contains the Wrstperson exclusive suYx -ay also appearing in detenay ‘I get up’); and its thirdperson plural counterpart is detenji (in which the appearance of the third-person plural suYx -ji overrides that of @-).5 Thus, not all morphosyntactic content of @deten can be seen as the contribution of an inXectional marking; nor could one say that Sora verb forms receive a second-person interpretation by default, since both the second- and third-person singular counterparts of @deten (deten ‘you (sg.)/s/he gets up’) lack any overt expression of person. At this juncture, proponents of incremental theories might propose that Sora possesses one or more phonologically empty person markers.
While the use of phonetically null aYxes is not new, it is nevertheless a questionable formal device, if only because the putative distribution of such aYxes is hard to demonstrate empirically. Often, zero aYxes arise in a Structuralist implication on analogy with the distribution of one or more overt aYxes with comparable but contrastive meaning. The incremental position in general implies that any content found in a word beyond the lexical meaning of the root is added either through a discrete operation with no phonological eVect (Steele 1995) or through the concatenation of a phonetically null but contentful aYx at some morpheme boundary. Taken to its logical conclusion, this move engenders either a large population of homophonous null aYxes or a potentially long derivation of string-vacuous rule applications. In either case, the argument is developed theory-internally, and it is therefore unfalsiWable. In inferential–realizational theories, however, nothing so exotic as zero aYxes is needed to account for the Sora facts; instead, one need only assume that in the inXection of Sora verbs of the de type, there happens not to be any rule explicitly realizing the property ‘‘second person’’.6
Lexical theories also carry unwarranted implications about morphological form; in particular, they imply that inXectional markings are like lexically listed words in at least two ways. First, they imply that inXectional markings are inserted from the lexicon into phrase-structural nodes, and are therefore always linearly ordered with respect to the expressions with which they combine. Second, they imply that two types of relation may hold between an inXectional marking and a morphosyntactic property (or property set): an inXectional marking may express a particular morphosyntactic property (set), or it may be restricted to the context of a particular property (set). Neither of these implications is well motivated. First, inXectional
5 The fact that the suYxation of -be or -ji overrides the preWxation of @- shows that these aYxes are members of an ambiWxal position class (one whose members include both preWxes and suYxes); for discussion, see Stump (1993b, 2001: 133, 284f).
6 For a detailed analysis of Sora verb morphology in an inferential–realization framework, see Stump (2005).

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 390 1.9.2006 5:00pm

390 thomas stewart and gregory stump

Table 12.2 Affirmative paradigms of the Sora verb de ‘get up’

Singular Plural

1 2 3
1 incl 1 excl 2 3

e

e

Nonpast
de-te-n-ay de-te-n de-te-n de-te-n-be -de-te-n-ay -de-te-n de-te-n-ji

e

e

Past
de-le-n-ay de-le-n de-le-n de-le-n-be -de-le-n-ay -de-le-n de-le-n-ji

Source: Biligiri 1965: 232 ff

markings do not necessarily combine with other expressions in the same ways that words do. The pluralization of Somali d´ıbi ‘bull’, for example, is eVected by a prosodic inXectional marking (dib´ı ‘bulls’); representing the morphology of dib´ı as an aYxational structure is at fundamental odds with any observable evidence. Second, there is no empirical motivation for assuming that an inXectional marking must be seen as expressing one set of morphosyntactic properties but as selecting for some other set of such properties; instead, one may always simply assume that the only relation between an inXectional marking and a set of morphosyntactic properties is the relation of exponence. In the inXection of Swahili verbs, for instance, the default mark of negative polarity is a preWx ha-, as in hatutataka ‘we will not want’. In the inXection of negative past-tense verb forms, the default past-tense preWx li- is overridden by a special suYx ku-: tulitaka ‘we wanted’, but hatukutaka ‘we did not want’. Although one could certainly treat ku- as expressing past tense but selecting for a negative context, there is no evidence to favour this approach over the simpler approach treating ku- as an exponent of both past tense and negation. An inferential–realizational theory of morphology such as PFM is fully compatible with this simpler approach.
We conclude on the basis of these considerations that the most adequate theory of morphology is both inferential and realizational. Logically, a realizational theory requires an explicit account of the association of morphosyntactic properties with their exponents, and an inferential theory requires an explicit account of the principles regulating the ways in which morphological rules compete or combine in the deWnition of inXected forms; PFM furnishes both of these, as we show in sections 12.6–8. First, however, we discuss a third distinctive aspect of PFM, namely its theory of paradigms (sections 12.4 and 12.5).

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 391 1.9.2006 5:00pm
paradigm function morphology 391
12.4 Content Paradigms and Form
Paradigms
.........................................................................................................................................................................................
Paradigms participate in the deWnition of two diVerent grammatical domains (Stump 2002, to appear a; Ackerman and Stump 2004). On the one hand, a lexeme’s paradigm distinguishes the various ways in which it can enter into the deWnition of phrase structure. In the syntax of Latin, noun phrases and their heads are speciWed for three morphosyntactic properties: a property of gender (masculine, feminine or neuter), which is lexically stipulated for each noun lexeme, and thus invariant within a given noun paradigm; one of six properties of case, as listed in (1) below; and a property of number (singular or plural). The paradigm of a Latin noun therefore canonically contains twelve cells, one for each of the twelve sorts of N nodes into which it might be inserted. The masculine nominal lexeme am"icus ‘friend’, for example, provides the paradigm schematized in (1); each cell in this paradigm is schematized as the pairing of am"icus with a diVerent gender– case–number speciWcation. (The paradigm itself need not, of course, be listed lexically; it need only be accessible by projection from information speciWed in the lexeme’s entry.) Seen as a response to the needs of syntax, (1) constitutes a content paradigm.
(1) Content paradigm of the lexeme am¯ıcus ‘friend’ a. h am¯ıcus, {masc nom sg} i g. h am¯ıcus, {masc nom pl} i b. h am¯ıcus, {masc voc sg} i h. h am¯ıcus, {masc voc pl} i c. h am¯ıcus, {masc gen sg} i i. h am¯ıcus, {masc gen pl} i d. h am¯ıcus, {masc dat sg} i j. h am¯ıcus, {masc dat pl} i e. h am¯ıcus, {masc acc sg} i k. h am¯ıcus, {masc acc pl} i f. h am¯ıcus, {masc abl sg} i l. h am¯ıcus, {masc abl pl} i
Besides entering into the deWnition of phrase structure, paradigms participate in the deWnition of a language’s morphological forms. In a realizational theory of morphology, rules of inXection apply to the pairing of a root7 with a morphosyntactic property set; the paradigm of a Latin noun therefore provides an inventory of twelve such pairings, as in (2). Realization rules such as those in (4) apply to the pairings in (2) to determine the realizations listed in (3). Seen as a response to the needs of morphology, (2) constitutes a form-paradigm.
7 Here and below, we adhere to the following terminological usage: a word form is a synthetic realization of a cell in a paradigm; a stem is a morphological form which undergoes one or more morphological rules in the realization of a cell in a paradigm; and a lexeme’s root is its default stem.

Ramchand and Reiss / The Oxford Hand book of Linguistic Interfaces 12-Ramchand-chap12 Page Proof page 392 1.9.2006 5:00pm

392 thomas stewart and gregory stump

(2) Form paradigm of the root (3) Realizations of the cells in (1)

am¯ıc ‘friend’:

and (2)

a. h am¯ıc, {masc nom sg} i a. am¯ıcus

b. h am¯ıc, {masc voc sg} i

b. am¯ıce

c. h am¯ıc, {masc gen sg} i

c. am¯ıc¯ı

d. h am¯ıc, {masc dat sg} i

d. am¯ıco¯

e. h am¯ıc, {masc acc sg} i

e. am¯ıcum

f. h am¯ıc, {masc abl sg} i

f. am¯ıco¯

g. h am¯ıc, {masc nom pl} i

g. am¯ıc¯ı

h. h am¯ıc, {masc voc pl} i

h. am¯ıc¯ı

i. h am¯ıc, {masc gen pl} i

i. am¯ıco¯rum

j. h am¯ıc, {masc dat pl} i

j. am¯ıc¯ıs

k. h am¯ıc, {masc acc pl} i

k. am¯ıco¯s

l. h am¯ıc, {masc abl pl} i

l. am¯ıc¯ıs

(4) Some Latin morphological rules a. Stem-formation rule Where root R is a second-declension nominal, R’s thematized stem is Ru. b. Realization rules Where X is the thematized stem of a second-declension root R and R is an adjective or masculine noun, i. cell h R,{masc nom sg}i is realized as Xs; ii. cell h R,{masc voc sg}i is realized as Re; ...

There is, of course, a close connection between the content paradigms and its form paradigms of a language. In particular, each cell in a content-paradigm (i.e. each content cell) normally corresponds to a particular cell in a particular form paradigm (i.e. to a particular form cell); this form-cell is its form-correspondent. In general, the realization of a content cell is that of its form correspondent. Thus, because the content cell in (1a) has the form cell in (2a) as its form correspondent, they share the realization in (3a).
In the canonical case, there is an isomorphic relation between a language’s content and form paradigms: a lexeme L has a single root R, and for each morphosyntactic property set s with which L is paired in some cell hL, si of its content paradigm, the form correspondent of hL, si is hR, si (so that the realization of hL, si is that of hR, si). This isomorphic relation might be formulated as the rule of paradigm linkage in (5), in which ) is the form-correspondence operator.

(5) The universal default rule of paradigm linkage Where R is L’s root, hL, si ) hR, si
MorphologySyntaxMorphosyntactic PropertiesTheoriesExponence