Faculty - Legendre

Director of Graduate Admissions


Theoretical linguistics: syntax and interfaces between syntax, morphology, and phonology. Acquisition of Syntax. Role of optimization in formal models of variation; Optimality Theory in syntax and acquisition of syntax. Cross-linguistic variation in syntax and morphology (with a particular focus on French and Balkan languages). General cognitive architecture underlying the language faculty.


Ph.D., Linguistics, University of California - San Diego, 1987.

Since 1992 I have sought to develop a new approach to the theory of syntax employing the framework of Optimality Theory. This work addresses a number of the central issues in contemporary syntactic theory: the role of serial derivation, violability and optimization in grammar, the extent and nature of variation within and across languages, the modularity of syntax and the nature of its interfaces to phonology and semantics, the relation of morphology and syntax, the role of discourse and information structure in syntax, and the acquisition of syntax. This research has progressed by tackling a series of central questions that posed difficult theoretical challenges yet find natural answers within my perspective.

How does universal grammar arise from neural computation?

The central notion of grammar as conceived in my earliest work in Relational Grammar[refs] is that a grammatical sentence is one that satisfies a set of parallel, inviolable well-formedness constraints. This formed the

basis of a collaboration with Paul Smolensky who had developed general characterizations of neural computation as parallel satisfaction of violable well-formedness constraints — optimization. In 1990, under NSF funding, we developed the first grammar formalism derived from neural computation, Harmonic Grammar [refs]. We developed this optimization-based framework through the analysis of a problem in French syntax that I had intensively studied in earlier work: unaccusativity mismatches [refs]. Harmonic Grammar was then used by Smolensky in collaboration with another linguist, Alan Prince, to study problems in phonology rather than syntax. This work led them to develop Optimality Theory (OT) between 1991 and 1993. This theory extended Harmonic Grammar in ways important for the formal study of linguistic universals, while moving it further from its neural computational basis; syntax was not addressed. In OT, a structure is grammatical if it is optimal relative to its competing structures, in the sense of best satisfying the conflicting demands of a set of simultaneous, universal constraints with differing strengths. Cross-linguistic variation exists because different constraints may have different strengths in different languages. Given a set of constraints, it is possible to compute a factorial typology of languages which formally expresses the extent to which linguistic systems may vary.

How can the role of optimization in syntax be formalized?

My earlier work on Harmonic Grammar having suggested an important role for optimization in syntax, I set out to see whether the refined formalization of optimization provided by Optimality Theory could be applied to syntax. There are numerous new challenges that syntax provides for OT, which in 1992 was still in development for phonology by Prince & Smolensky. Smolensky and I collaborated again to extend OT to syntax; with one of my students, we gave the first conference presentation of work in OT syntax in February 1993. Building on earlier work of mine in case-based syntax [refs], we had set out to test the power of our conception of OT syntax in the cross-linguistic domain of case marking systems (accusative, ergative, and stative) and voice (passive, antipassive, and reversal). We showed that highly general and rampantly violated constraints on the mapping of argument structure to syntax (e.g., ‘agents are subjects’), when formalized in OT, entailed a small typology of possible languages which we were able to document empirically [refs].

A second crucial challenge to OT posed by syntax had to be addressed next: How can it be that certain meanings are ineffable — inexpressible — in some languages, if the best alternative expression of a meaning is, by definition, grammatical? (For example, one cannot form a question that has two question words like ‘What did he send where?’ in Italian or Irish.) A final collaborative project with Smolensky, started in 1993, was the first to address this important problem. In joint work with several of my students under NSF funding, we argued that correctly incorporating into syntax the two types of constraint families active in OT phonology — markedness constraints and input-output faithfulness constraints — provided a straightforward answer to the ineffability question. An input (a meaning) is ineffable when faithfulness is outranked by markedness, yielding as optimal an output whose meaning is distinct from the one intended in the input [refs]. This completely new idea was presented at a number of forums in 1994 and 1995, including the MIT conference on Optimality in Syntax, and has since been adopted by many OT syntacticians (P. Ackema, A. Neeleman, E.Bakovic, E. Keer, C. Wilson, and others).

The work in which we investigated ineffability is in fact a novel, general OT theory of extraction phenomena (in particular wh- and related movement), providing an in-depth account of over a dozen constructions in three typologically diverse languages (English, Bulgarian, Chinese). This theory is the only one I’m aware of that manages to derive ‘relativized minimality’ effects in extraction phenomena from the OT-interaction of simple violable constraints, none of which involve minimality or relativization [refs]. The paper [ref: see previous #24], in press from 1995 to 1998, provides a simple, general characterization of the theory, a high-level, encapsulated version of the grammars for English, Bulgarian, and Chinese, and a wide-ranging discussion of the theory’s typological predictions.

These projects established the groundwork of a general approach to syntax in OT. My work since 1995 has focused on explaining further syntactic and interface phenomena as well as testing OT as a model of acquisition of syntax.

What is the formal architecture of the grammar? Is the relation between components of the grammar a derivational/sequential one or a parallel one?

Investigating this important issue requires examining the interface between components and focusing on key phenomena which straddle several components, like clitics (syntax, morphology, and prosody). Clitics are phonologically weak elements (such as Romance object pronouns and Balkan tense auxiliaries) which, traditionally, have been taken to play a role in the syntax. Yet, they are subject to positional restrictions unheard of for syntactic categories, including prosodically-based ones. In a series of studies of three different Balkan languages with three different clitic patterns [refs]culminating in a recently completed synthesis [ref: see previous #51], I have argued that the positioning of clitics is largely the result of the interaction of two basic violable interface alignment constraints, one that favors aligning the phonological (PF) realization of a feature as near as possible to the edge of a syntactic phrase domain, and the other disfavoring positioning it at the very edge of an intonational phrase domain. Other constraints, syntactic or prosodic, may also affect their positioning with the consequence that the range of cross-linguistic variation reduces to two broad constraint rankings: syntax outranking PF, and PF outranking syntax. This work, as well as the extension described in the next paragraph, entails that the relation between components is a parallel, rather than a derivational, one.

What is the status of (inflectional) morphology in the grammar? Is it subsumed under syntax? Is it a separate component, sandwiched between syntax and phonology? Is it subsumed under the lexicon? Is it none of these?

My work on clitics since 1996 and its recent extension to functional categories in general (including Tense and Agreement) [refs] have confronted this contentious issue head-on. A detailed analysis of the syntactic behavior of clitic auxiliaries and clitic pronouns in Balkan languages has convinced me that they are morphological categories — phrasal affixes — that are syntactically inert, contrary to the belief held by many generative syntacticians. A detailed study of verb-second phenomena in Breton [refs] and Basque [refs] and the fact that these phenomena are often masked that is, demonstrably present but not overtly detectable — have also convinced me that finiteness features (Tense and Agreement) are subject to the same violable interface alignment constraints as clitics. In particular, the verb second phenomenon is blind to the syntactic or morphological status of auxiliaries. I have argued that a unified analysis is only possible if verb second is construed as resulting from the interaction of violable PF constraints. Over all, this empirically rich domain of investigation has led me to side with some prominent morphologists (e.g. S. Anderson, G. Stump, A. Spencer) and adopt a realizational view of morphology. Morphology is a set of well-formedness constraints at the syntax-phonology interface, each describing some modification of an existing lexical item that relates it to other forms. To some extent, this work grew out of my own earlier work on secondary predication and agreement in French, conducted within the more traditional (i.e., syntactic) perspective on inflectional categories, which proved to be less than satisfactory [refs].

I have also reexamined the alleged evidence that clitic pronouns are controlled by syntactic principles and argued that their syntactic behavior reduces to the fact that they are agreement markers in an agreement relation with a syntactic category, whether that relation is overt (clitic doubling) or not. Clitic climbing, on the other hand, is only a subcase of a larger phenomenon of syntactic domain extension and alignment of functional features with main clause Tense [refs].

On-going work explores the variation in subject clitic doubling across different registers of a single language, French, as well as across languages. This is part of a large typological study of clitic systems and an attempt to characterize the formal nature of intra- vs. cross-linguistic variation.

What is the status of information structure (discourse) in the grammar, and how does it relate to ‘optionality’?

Many word order effects are tied to discourse factors (e.g., focus) and OT provides a straightforward way to formalize the syntax-discourse relation. In my recent work on Stylistic Inversion in French wh-questions (Où est allé Jean? ‘where did John go?’) I have proposed that the postverbal position of the subject is a consequence of its discourse status: +identifiable, -active [21]. In particular, I have argued that the traditional discourse status notions of focus and topic, must be replaced by these finer-grained features specified in the input to optimization.

Discourse factors and faithfulness to discourse features in the input are relevant to masking verb-second phenomena as well, as shown in my study of Breton and Basque [refs]. In fact, discourse factors, all too often overlooked by theoretical syntacticians, have been crucial to my theory of variation in OT syntax since the very first work on voice alternations in 1993 [31].

Can the development and variability of early syntax be precisely accounted for via the same principles governing adult syntax?

OT encompasses a comprehensive formal theory of constraint interaction as well as an explicit theory of learning (Tesar and Smolensky, 1998). Given knowledge of grammar, the task of the learner in OT is to determine the constraint ranking which singles out the target language from the space of possible languages. In 1997, I started a collaborative project in first-language acquisition of syntactic structure with a student of mine and a visiting researcher, Anne Vainikka, in order to test the hypothesis that this process occurs over time via re-ranking of universal constraints. This work focuses on the very early development of inflection and related properties in a number of languages (roughly between the ages of 1½ and 2½ years of age). The first step in this project, funded by an NSF grant (Learning and Intelligent Systems), was a methodological one: the design of a cross-linguistically valid, independent measure of different stages of development in syntactic production which avoids the problems that come from lumping together several stages of development. The measure, presented in [51], is called the Predominant Length of Utterance (to highlight its relation with another, morpheme-based, measure commonly used in developmental studies but not particularly useful in syntactic acquisition, Brown’s Mean Length of Utterance).

Armed with our new measure of developmental stages, and focusing initially on child French, we set out to explain, among other things, two empirical generalizations we uncovered in the development of tense and agreement. One is that (at least in French) acquisition of tense suffers a dip as agreement emerges. We propose that tense and agreement compete for realization in a single, underspecified, functional projection at the relevant stage. The other novel observation is that the relative distribution of finite and non-finite verbal forms through the course of the ‘optional infinitive stage’ (Wexler, 1994) is not as random as this global characterization would lead us to expect. At every fine-grained stage of development, the relative distribution of finite and non-finite verbal forms can, in fact, be quantified and correlated with other grammatical properties, revealing a pattern which is not one of unrestricted optionality. Previous formal models of acquisition shed no light on the frequency with which non-finite forms are used; nor do they provide any obvious means by which to express, let alone explain, the observed percentages and correlations.

Our OT theory of variation in child syntactic production [21] posits a developing partial ordering of faithfulness constraints with respect to a fixed hierarchy of markedness constraints. (Partial orderings of violable constraints have previously been used to explain variation in phonology). The partial order is formally a set of rankings, where each yields a potentially different output. The proportion of each output type in the set predicts the proportions of attested forms in the production data. Other striking properties of child French that distinguish it from adult French, the occurrence of null subjects and postverbal subjects, are also subsumed under a general analysis of child French [21]. Subsequent work will address the development of inflection in morphologically-rich languages. With the help of native speakers drawn from our graduate and undergraduate population, we have, in fact, started the analysis of relevant production data in Catalan, Hungarian, and Polish. Experimental work on syntactic perception is also planned to empirically explore general OT proposals concerning the relation between child production and perception.

(Numbers are referenced in the Research Summary)

Books

1. Smolensky, P. and G. Legendre. In progress. The Harmonic Mind (Anticipated date of completion: Fall 2002)

2. Legendre, G., J. Grimshaw, & S. Vikner (eds.). 2001. Optimality-theoretic Syntax. MIT Press.

Articles

4. Hale, J. and G. Legendre. In progress. Minimal Links, Remnant Movement, and (Non)-derivational Grammar.

5. Legendre, G, A. Vainikka, P. Hagstrom, and M. Todorova. 2002/Under review. Partial Constraint Ordering in Child French Syntax.