languages or within a single language a whole host of morphological, lexical and syntactic ..... Now, the analytic passive doesn't have this freedom. Instead, the ...
1
Syntax as an exponent of morphological features Andrew Spencer and Louisa Sadler, Department of Language and Linguistics University of Essex Colchester, CO4 3SQ UK {spena,louisa}@essex.ac.uk
1. INTRODUCTION...............................................................................................................................3 2. ANALYTIC PERFECTIVE PASSIVES IN LATIN ........................................................................5 3. F-STRUCTURES IN CLASSICAL LFG ..........................................................................................8 4. LATIN PERFECTIVE PASSIVES AS TRUE PERIPHRASTIC CONSTRUCTIONS - A WRONG SOLUTION...........................................................................................................................10 5. BÖRJARS ET AL. 1997 ON LATIN PERIPHRASIS ...................................................................15 6. M-FEATURES AND S-FEATURES: THE MORPHOSYNTACTIC PROJECTION INDEX .18 7. RULE OF REFERRAL SOLUTION TO THE LATIN PERIPHRASIS .....................................24 8. CONCLUSIONS ...............................................................................................................................33
2
Syntax as an exponent of morphological features*
Abstract
We examine the periphrastic passive construction in Latin, in which a part of the verb paradigm is expressed by an auxiliary/copular verb ‘to be’ with the perfective passive participle, having the syntax of a predicative adjective construction. We show that an analysis of this construction offered by Börjars, Vincent and Chapman 1997 within the LFG framework suffers from conceptual and technical flaws. We show that the Latin data require us to draw a crucial distinction between m(orphological)-features and s(yntatic)-features. The syntactic features are represented in LFG functional structures, while the morphological features belong to a separate, morphological, level of structure. The analysis is couched within Stump’s (in press) Paradigm Function Morphology and appeals to the notion of a rule of referral within the morphological paradigm, defined over m-features, realizing the perfective passive features as a syntactic construction. We show that the referral must be stated over m-features, that is, at the level of the morphological verbal paradigm, because of the existence of deponents. In those verbs the form is passive but the meaning is active. However, the perfective tenses are formed analytically, in keeping with the ‘passive’ morphological marking. The analysis shows how a realizational theory of morphology of the kind which countenances rules of referral can throw light on questions of the morphologysyntax interface.
*
Versions or parts of this paper have been read at a meeting of the ESRC-funded
seminar ‘Challenges for Inflectional Description - 6’, University of Surrey, 24 June 1998, the Linguistics Association of Great Britain (University of Manchester, April 1999) and the Second Mediterranean Morphology Meeting, University of Malta.
3
1.
Introduction
In this paper we investigate a selection of issues in the morphology-syntax interface. This has been the locus of intense research activity in recent years particularly within lexicalist theories of grammar such as LFG. A central question is how it is that across languages or within a single language a whole host of morphological, lexical and syntactic means can be deployed to express essentially the same set of meanings or functions. One very specific example of this is seen when very similar (or even identical) grammatical meanings/functions are expressed now by inflected morphological word forms and now by means of small syntactic constructions, in other words, when a single set of grammatical properties receives synthetic and analytic expression.
We consider one such system in some detail, the perfective passive subparadigm of Latin verbs, discussed recently within an LFG framework by Börjars, Vincent and Chapman 1997. Our overall approach to morphology is that of the word-and-paradigm school (cf. Ackerman and Webelhuth 1998), specifically the model of Paradigm Function Morphology developed in Stump (in press). In such a model, there is no appeal to morphemes as signs, or lexical entries with their own properties which are projected in morphosyntactic representations. Rather, morphological paradigms are generated by realizational rules (or their formal equivalent, such as statements of inheritance). Stump refers to this class of theories as ‘inferential-realizational’ and argues on a host of empirical and conceptual grounds that this class of theories is superior to alternatives, especially those that appeal to the classical morpheme concept.
Morphological theory has devoted very little attention to the way in which a wordand-paradigm morphology might interface with syntactic structure, however. The Latin periphrastic construction gives us a very useful window for examining this interface. We show that systems such as the Latin require us at the minimum to distinguish two sorts of features, morphological features (m-features) and syntactic features (s-features). In the ideal case the two coincide, in the sense that a single morphological word form will have a clear and unambiguous syntactic use and that
4
syntactic use will be invariably be expressed by means of the given morphological form. However, as recent morphological research has emphasized, this one-one mapping is an ideal which is often breached (Aronoff 1994, Beard 1995, Stump in press). The Latin case is a particularly good example of the need for ‘separationism’ between morphology and syntax. On the other hand, to capture the fact that there often is, indeed, a reasonably straightforward match between the morphological and the syntactic side of morphosyntax we introduce a notational device to allow us to state that the trivial mapping is the default mapping: s-features and the m-features which realize them can be thought of as a single feature with different projection indices (though other representational devices for coding the same intuition are conceivable). In the default situation the m-indexation and the s-indexation coincide, but this default can be overridden by various kinds of morphosyntactic mismatch. It is the existence of such mismatches which motivate the distinction between m- and s-features.
Given this morphological architecture, we turn to the Latin analytic passive forms. An initially attractive way of looking at these forms is to think of them as syntactic constructions which just happen to express more or less the same meanings as the missing morphological forms, and which are therefore used in their stead. This can only work if we say that the morphological verb paradigms totally lack the perfective passive forms. However, we show that this leads to untenable results. The problem is that there are deponent verbs, which are passive in form but active in function/meaning. In the perfective aspect these therefore have the analytic construction, but this can’t be related to the passive meaning of the passive participle, because the forms are active in meaning. As we show in Sectin 4, this entails that the passive perfective forms are part of the paradigm of the deponent verb, and this in turn means that the corresponding perfective passives of non-deponents are also part of the paradigm. We are thus left with the key analytic question: how can the grammar be so organized as to ensure that the analytic construction has the morphosyntax of a predicative adjective construction, while simultaneously realizing the perfective passive features of a verb paradigm? Our answer is to appeal to the same device employed by Stump (in press) for the similar situation of the ‘periphrastic future’ of Sanskrit. We propose a rule of referral, defined, crucially, over m-features, which tells us to realize the requisite m-feature combinations as a predicative adjective + copula
5
construction (using the perfective passive participle form of the verb). This solution obviates a number of problems found in the account of Börjars et al. 1997.
The paper is organized as follows. In section 2 we sketch the essential facts of the Latin system. In section 3 we illustrate the LFG notion of f-structure, a level of syntactic projection which permits us a unified statement of grammatical functions independently of their formal realization. In section 4 we introduce and motivate the separation of m-features and s-features, and in section 5 sketch the idea of projection indices. Section 6 completes the discussion of the architectural assumptions we are making by discussion the need for sets of conditions defining what a well-formed paradigm is in both syntax and morphology.
In section 7 we provide a thumbnail sketch of the kind of analysis that we would expect to see if we tried to take seriously the idea that the perfective passive participle form, in itself, meant ‘perfective’ and ‘passive’ and that this should be used to drive the periphrastic construction. Despite a number of attractions, we show that this solution fails because it is unable to deal with the deponent verbs. Our own solution, which appeals to a rule of referral, is given in section 8. Section 9 presents summary conclusions. 2.
Analytic perfective passives in Latin
As discussed in (Börjars et al. 1997:167f), Latin has two verbal aspects, imperfective and, for non-present tenses, perfective. It has an active and passive voice. In the imperfective aspect forms the passive is expressed synthetically, but in the perfective tense series we find an analytic construction formed from the copular verb sum ‘to be’ and the perfective passive participle (PPP) formed in -t-. This is illustrated in (1):
6
(1)
Latin verb forms
laudo ‘I praise’ IMPERFECTIVE
active
passive
present
laudat
laudatur
past
laudabat
laudabatur
future
laudabit
laudabitur
PERFECTIVE
active
passive
present
laudavit
laudatus/a/um est
past
laudaverat
laudatus/a/um erat
future
laudaverit
laudatus/a/um erit
The participle is morphosyntactically an adjective. Thus, while finite verbs agree with the subject in Person/Number features, the analytic perfective passive forms agree in Number/Gender (but not person), exactly as predicative (and attributive) adjectives do. This is illustrated in (2), where we see 3rd sg. forms:
(2)
Agreement in perfective forms Pf. Act.
Pf. Pass. (Fem. subj.)
1sg
laudavi
laudata sum
3sg
laudavit
laudata est
3pl
laudaverunt
laudatae sunt
In (3-5) we illustrate the morphosyntax of the participial constructions in comparison to normal adjectives:
(3)
adjective agreement (all in the Nominative case):
roman-us sum
roman-a es
roman-ae sunt
roman-MASC.SG am
roman-FEM.SG art
roman-FEM.PL are
‘I (masc.) am Roman’
‘thou (fem.) art Roman’
‘they (fem.) are Roman’
7
(4)
adjective construction: Clodia romana est C. Roman.FEM.NOM.SG is ‘Clodia is Roman’
(5)
participial passive construction Clodia laudata est C. praised.FEM.NOM.SG is ‘Clodia was/has been praised’
One further important feature of Latin inflection is the existence of deponent and semi-deponent verbs. Deponents are verbs with active syntax and active ‘meaning’ but which have the form of passive verbs. The semi-deponents are active in form in the imperfective tense series but take the deponent, passive, form in the perfective aspect. In (6) we see a sample partial paradigm for a deponent verb and in (7) we see a semi-deponent:
(6)
(7)
Deponent verb: loquor ‘I speak’ IMPERFECTIVE
PERFECTIVE
present
loquitur
locutus/a est
past
loquebatur
locutus/a erat
future
loquar
locutus/a erit
Semi-deponent verbs: gaudeo ‘I rejoice’ IMPERFECTIVE
PERFECTIVE
present
gaudet
gavisus/a est
past
gaudebat
gavisus/a erat
future
gaudebit
gavisus/a erit
Deponent and semi-deponent verbs are largely intransitive. Some, however, take Accusative marked direct objects (totam causam oblitus est ‘he forgot the whole case (Acc)’ from obliviscor ‘forget’ and a number take oblique case-marked complements
8
such as gratulor ‘congratulate + Dative’, utor ‘use + Ablative’. Very occasionally a deponent can be used with the function and interpretation of a genuine passive (!), a fact which, while interesting, will not concern us. 3.
F-structures in classical LFG
Lexical Functional Grammar (Bresnan, in press) posits two levels of surface syntactic representation, c-structure which expresses information about constituent structure and syntactic word class, and f-structure which captures information about grammatical relations and such semantically interpretable functional features as definiteness, tense and so on. Together, these levels of representation capture the observed wide variability of external surface form (exponence) together with the largely invariant or universal aspects of syntactic structure. It is an important aspect of LFG that it is possible to express grammatical functions independently of exponence. Thus, whether the subjects and objects of a clause are expressed by word order, agreement, case marking, clitics or whatever (or any combination of these) the fstructure will remain relatively constant across languages and across constructions. This level of representation therefore makes it possible to state generalizations about functional organization which cut across morphosyntactic realization. For instance, the f-structure shown in (8) will be applicable to any language that makes a tense distinction and which has pronominals, whether full pronouns, clitics or affixal (incorporated) pronominals, which distinguish person and number:
9
(8)
f-structure for ‘I saw them’
PRED TENSE SUBJ OBJ
' see < (SUBJ), (OBJ) >' PAST PRED ' PRO' PERS 1 NUM SG PRED ' PRO' PERS 3 NUM PL
In classical LFG analyses the f-structure completely lacks information proper to cstructure, namely, categorial features (such as [±N,±V] or their equivalent), and the independent representation of ‘functional’ categories such as DET, AUX, SPEC, and so forth. The distinction between c-structure features and f-structure features turns out not to be quite so clear-cut, however. Consider, for instance, the analysis in (9) of ‘was persuaded (to go)’ (Kaplan and Bresnan 1982:224):
(9)
persuaded: V, (↑PARTICIPLE) = PASSIVE (↑PRED) = ‘PERSUADE
