There is an ongoing debate in generative linguistics on what should be considered a "word", and what component of language is responsible for its formation. Words have been defined as phonological units, as the basic elements in the lexicon, as the output of a morphological component or as the indivisible terminal elements of syntactic structure. One of the central issues in this debate is the relation between morphology and syntax, which have often been characterized as two distinct yet parallel components of grammar responsible for the formation of linguistic objects. More recent studies have questioned whether such a distinction should be drawn at all.

Since Chomsky (1970) and the acceptance of the Lexicalist Hypothesis, which claims that syntax does not have access to the internal form of words, a view of the architecture of grammar has emerged in which two distinct and autonomous components of morphology and syntax coexist. Since the Lexicalist Hypothesis asserts that syntax cannot combine morphemes into words, an independent and separate mechanism had to be posited. Past approaches in the field have often argued that morphology is a distinguishable subtheory of language with principles and vocabulary items that are distinct from the principles and atoms in syntax or phonology. Hence, in the generative framework, morphology or a morphological lexicon is traditionally treated as the component of word-formation. The resulting "morphological words" behave as an atomic entity in syntax and syntactic principles cannot manipulate their internal structures. Syntactic objects, however, are not syntactically atomic and their parts are visible to the principles of syntax. This split in behavior based on the component in which an element is formed has led to the standard dichotomy between words and phrases, where words are created in a morphology component while phrases are formed by operations in syntax.

The strict word/phrase division, however, does not hold under scrutiny. Many linguistic phenomena raise challenges for this view and cross-linguistic studies have shown that certain linguistic constructions seem to cut across the word/phrase boundary, with phrases that behave as lexical or morphological units (e.g., deverbal nominalization) and morphological words that have the meanings usually related to clauses or phrases (e.g., in polysynthetic languages). The dual properties of linguistic objects that behave as "words" in some contexts and as "phrases" in other situations have blurred the boundary between what constitutes a word and what includes a phrase and raise a challenge for lexicalist models that insist in the formation of words and phrases in two distinct components. The linguistic variation with respect to the concept of "word" has resulted in an attempt to redefine this notion, and to rethink the interaction between morphology and syntax in the context of the duality of linguistic objects and the mismatches between meaning and surface realization.

This dissertation investigates verbal constructions across several languages, in particular valency changing phenomena in Persian and Eastern Armenian, and argues to eliminate the strict separation between components responsible for the formation of words and phrases. It proposes a notion of "word" defined as a level in syntax, the internal structure of which is visible to syntactic principles. The model proposed shares the main ideas of recent syntactic approaches such as Distributed Morphology (Halle and Marantz 1993, Marantz 1997) and the various systems developed in Travis (1991), Hale and Keyser (1993) and Vergnaud (2000), and argues for a single computational domain in grammar which is responsible for the formation of all linguistic objects. By investigating the internal structure of verbal predicates, the thesis isolates the atoms used to encode meaning in natural language and formalizes a theory of the morphology-syntax interface.

This chapter presents some of the past research on the interface of morphology and syntax that has set the stge for this dissertation. I discuss several syntactic approaches such as Baker (1988), Hale and Keyser (1993) and Halle and Marantz (1993). In addition, I discuss Chomsky (1970) and Di Sciullo and Williams (1987), which have often been treated as works that defend the lexicalist apporach to predicat-formation and demonstrate that the alternative approach would not be able to account for the data observed. However, the discussion presents several points from these two writings that counter these conclusions. The chapter also reviews the main theoretical issues that arise in an investigation of the notion of "word", introduces the basic proposal of this dissertation and the architecture of grammar that emerges as a result.

Causative constructions refer to complex predicates formed by the combination of a causative event with an underlying predicate. The addition of the causative verbal element also adds a new participant (a causer), which initiates or controls the event of the underlying predicate. Typologically, several diverse methods are used for forming causatives, and the properties of these constructions vary from one language to another. Some causative constructions appear as a single morphological word, while others are formed periphrastically by adding a causative verb to the base predicate. Structurally, certain causative predicates behave as mono-clausal constructions denoting direct causation, while others display properties of a biclausal construction and represent an indirect causation of the underlying event. There exist, however, certain cross-linguistic parameters marking the correlation between the mechanism used for forming a causative and the resulting meaning of the predicate (Comrie 1989, Dixon 2000). This chapter closely examines the causative constructions in Eastern Armenian and tries to provide an explanation for some of these correlations.

In the literature on causative constructions, morphologically formed causatives have traditionally been analyzed as composed within the morphological or lexical component, while the creation of periphrastic or analytic causatives is associated with the syntactic domain. The surface form and the meaning or structural properties of causative constructions, however, do not always coincide. Hence, it is not the case that a morphological causative would always correspond to a monoclausal structure or that an analytic causative should always be interpreted as an indirect causation. It has been argued, in fact, that the morphological causatives in languages such as Japanese (Harley 1995) or Malagasy (Travis 1999) behave like predicates that contain two eventive verbs; and although Urdu permissive constructions and Italian causatives appear as two independent verbs, they behave as a simple predicate (Butt 1995 for Urdu and Di Sciullo and Williams 1987 for Italian, among others).

This chapter begins with an introduction to causative constructions in Japanese, which clearly demonstrate mismatches between the surface form and the meaning of these complex predicates since the morphological causative in this language sometimes shows biclausal properties. In this chapter, I argue that such mismatches between form and meaning raise an important problem for the strict division between the component responsible for the formation of morphological words and the component responsible for creating phrases and sentences. I will then present data from Eastern Armenian. An investigation of causatives in this language provides some insight into the distinct structures and properties of these constructions since Eastern Armenian is `well-behaved' in that it shows a direct correlation between surface realization and underlying structure. A syntactic analysis is developed that accounts for the different causative types in the languages discussed and captures the distinct surface realizations of causative constructions based on the resulting syntactic configuration. This approach does not need to posit several levels or subcomponents within the lexicon and provides a unified solution for different causative types, while providing an account for their distinct properties.

The chapter introduces data from Japanese causatives which show a clear mismatch between form and meaning, and presents the distribution of the morphological and analytic causatives in Eastern Armenian. By studying the syntactic properties of the two constructions, it is concluded that they are, in fact, distinct clausal structures and that the differences can be explained if the morphological causative contains only one "event" or light verb v in its structure, while the analytic causative includes two distinct events or light verbs. I provide a uniform analysis for the two causative types within syntax, and ague that the boundary between the morphological and analytic predicate subdomains can be represented based on the resulting syntactic structure. The analysis maintains, however, that the two causative predicates have very distinct characteristics. This is shown by an investigation of the semantic properties of the two constructions, such as the agency of the causee and the availability of idiosyncratic meaning, which shows that a syntactic analysis of complex predicate formation, which takes into account the number of events and the presence of the external argument, can capture the readings obtained and hence, there is no need to posit two distinct components of verb-formation. I then apply the proposed analysis to the Japanese data and propose an analysis using "derivation by phase" (Chomsky 1999) to provide an account for the distinct surface realizations of similar causative constructions in Eastern Armenian and Japanese based on the interface relations. The final section summarizes the discussion and provides a classification of causative types that is suggested by the proposed analysis.

The preceding chapter investigated one type of form and meaning mismatch, in which the same surface realization seems to display distinct syntactic and semantic properties indicating that they belong to different structural configurations. This was the case, for instance, in Japanese where the same morpheme -sase forms three distinct causative constructions. In this chapter, we will examine a second type of mismatch which arises in languages forming complex verbal constructions. In these languages, verbal predicates are created by combining a light verb element with another predicate which usually provides the substantive information to the complex construction. Examples of such complex predicates are light verb constructions in Indo-Iranian languages, particle verb constructions in Hungarian, and serial verbs. These verbal constructions always form complex predicates containing more than one grammatical element, each of which contributes part of the information that is ordinarily associated with a single head or morphological word in other languages. The challenge these complex verbal predicates raise for the word/phrase dichotomy assumed in lexicalist approaches is twofold: If a strict division between the morphological and the syntactic components is assumed, the theory will have difficulties in accounting for the observation that the same information can be represented as a single word in one language or as several independent words (i.e., phrases) in another. If all verbal meaning features are already available in the lexical entry, then how does the system capture the fact that different languages opt to project the lexical information into distinct components of grammar? Furthermore, complex verbal predicates often display dual properties sometimes functioning as a lexical item and at other times behaving as a syntactic object. This, of course, raises a dilemma if the theory clearly distinguishes between linguistic objects formed in the lexicon and those created in syntax.

Persian is an Iranian language that employs light verb constructions. The language uses a productive method of verb formation which creates complex predicates consisting of a preverbal element and a light verb. The preverbal element can be a noun, an adjective, an adverb or a preposition phrase, which combines with a verb to form a single syntactic predicate. Persian light verb constructions (also known as compound verbs or complex verbs) are directly relevant to the discussion of the word/phrase dichotomy, since they behave as single verbal predicates but are represented as two distinct morphophonological words. In addition, these constructions show an overt decomposition of the verb as proposed in many generative frameworks. Determining the syntactic and semantic contributions of these two components lies at the center of the investigation of compound verbs. In the literature on these constructions in Persian, it has been suggested that light verbs are semantically empty and it is the preverbal nominal element that lends its arguments to the complex verb (Mahammad and Karimi 1992). Others have argued that light verbs contribute aspectual information but not argument structure (Karimi-Doostan 1997). Barjasteh (1983), Vahedi-Langrudi (1996), Dabir-Moghaddam (1997) and Ghomeshi (1994) treat the preverbal element as an argument of the verbal component. Karimi (1997), however, has argued that each component of the complex verb contributes a thematic structure; these two structures undergo a semantic fusion after incorporating at LF.

Another issue that is directly related to the current discussion is the dual behavior of Persian complex predicates as lexical and syntactic elements. These verbal constructions undergo nominalization, adjectival formation and have a single primary stress, which has led some researchers to suggest that they are lexical units. On the other hand, complex predicates are visible to syntactic and morphological processes: The components of these verbs can often be separated by negation and inflectional affixes, auxiliaries, modals and emphatic elements. Furthermore, certain preverbal elements can act as full-fledged noun phrases, since they may be modified, gapped or relativized.

In this chapter, I investigate the causative/inchoative alternation in Persian and propose an analysis in which the arguments are not projected from the lexicon but formed compositionally by combining the basic components of the complex predicate in syntax. I suggest an analysis based on a syntactic decomposition of the verbal construction, following ideas developed in Marantz (1997), Chomsky (1995), Vergnaud (2000) and Hale and Keyser (1993), whereby the verbal structure is formed by the conjunction of the root element and functional components. I argue that the substantive aspects of the predicate are contributed by the preverbal element, while the event information (such as causation) is carried by the light verb. In addition, Aspect is closely related to the choice of the light verb. It is argued that aspectual properties and the interpretations of the arguments can be derived from the resulting syntactic configuration (Borer 1994, Ritter 1998). The analysis proposed can also capture the dual nature of complex predicates in Persian: Since word formation takes place in syntax, the fact that the internal structure of a complex verb is visible to syntactic and morphological processes follows naturally from this analysis. Nominalizations and adjectival formations, which have been used to argue for the lexical behavior of complex predicates, are a process of word formation and are derived in syntax (see Van Hout 1998, among others). If word-formation is not confined to the lexicon, complex predicates may undergo nominalization in syntax, without having to be treated as lexical or X^o units.

In Chapter 3, an investigation of the causative/inchoative alternation verbs in Persian allowed us to determine the primitive elements of the verbal predicate. These primitive units consist of roots, categorial features (e.g., adj) and verbal functional elements (e.g., v{become}, v{cause}, tense, aspect), which combine to form verbal predicates in syntax. Hence, these primitive elements are abstract categories defined by universal features, which combine following syntactic principles such as head-to-head movement to create new verbal predicates. In the theory developed in Distributed Morphology, the Lexicon contains only these primitive aspects of syntactic categories, which are used to construct phrase-markers following syntactic operations. The idea of the grammatical lexicon presented in this dissertation, however, differs from the view in Distributed Morphology since it is argued that the basic lexical items do not necessarily correspond to the primitive syntactic elements, and this is in fact a major source of variation in the world's languages.

The study of the light verb constructions in Persian resulted in a proposal based on the decomposition of the verbal predicate into its primitive elements representing roots, and categorial and functional entities. It was argued that these elements are all that is needed to compositionally form the verbal constructions within the computational domain. Hence, for Persian, the grammatical lexicon consists solely of these primitive syntactic units and does not contain any morphophonological expressions with predetermined argument structures. The meaning of the verb is derived from the final structure constructed. The primitive elements of grammar are argued to be universal in the sense that verbs in all languages are formed using various combinations of these syntactic features. Hence, an inchoative verb is always formed by combining the root, the categorial feature for adjective and the inner light verb projection v{become}. The causative version of the same verb would then include in addition the outer light verb projection for v{cause}. Recall that each functional category v also projects a specifier or argument position, which accounts for the valency distinction between the inchoative and causative variants, and an aspectual projection.

An examination of the causative/inchoative alternation verbs in Eastern Armenian, however, suggests that languages differ with respect to the size or structure of the items stored in their grammatical lexicons. In particular, I will show that, in order to account for the Eastern Armenian data, a distinction needs to be made between the notions of primitive elements of grammar and basic lexical items. Although inchoative verbs and their causative alternants have the same syntactic structure and meaning in both Persian and Eastern Armenian, formed by combining the primitive elements, the lexical item in Eastern Armenian can actually be composed of an association between several primitive elements. I will show that there exist two distinct classes of alternating verbs in Eastern Armenian with different properties. In particular, I claim that the causative alternants of verbs such as dry are formed by combining the various primitive elements in syntax, whereas verbs such as break are inherent causatives. I argue that the first group of alternating verbs are stored as an adjectival element in the grammatical lexicon, while the members of the second group are listed as an unordered set of associations corresponding to a full verbal structure equivalent to vP. The distinction between the basic lexical items of verbal predicates captures the different morphological, syntactic and semantic properties of the two causative constructions in Eastern Armenian.

The first Section investigates the transitivity alternation verbs in Eastern Armenian. The Armenian data provide support for the distinct classification of break vs. dry verbs proposed by Hale and Keyser. This section argues that these verbs do not constitute a uniform class in this language, and their difference is related to the lexical item in each case. The discussion leads to a distinction made between the notion of basic lexical item and primitive elements of grammar, whereby the lexical item may correspond to a structural association among various primitive syntactic features. The proposed analysis is compared to analyses of similar Greek data presented in Embick (1998) and argues that without the concept of basic lexical item the syntactic theories of word formation will fail to account for the two categories of alternation verbs in Eastern Armenian. I then contrast the causative/inchoative alternation verbs in Eastern Armenian to the ones in Persian discussed in the previous Chapter. It is shown that the proposed model can account for the similarities and differences of these verb types across languages if the basic lexical item in each language and the node determining the PF-phase are treated as parameters. Thus, the primitive elements of grammar are universal features that combine to construct verbal predicates, but a language's lexical items and the syntactic node at which the structure is delivered to the PF component can vary from one language to another. The combination of these two factors, therefore, contributes to linguistic variation in verb formation.

The previous chapters have concentrated on isolating the primitive elements of grammar that combine to form verbal predicates in Persian and Eastern Armenian. The discussion focused on the formation of the verb and little was said about the projection of the arguments in the verb phrase. The current chapter is concerned with the status of the nominal elements in the verbal predicates, with focus on the direct object and the nominal preverbal element in light verb constructions.

In most current syntactic theories, the verbal entry in the lexicon is assumed to include all the information needed for the projection of the arguments in the syntactic structure. Though there have been discussions over the type of information contained in the lexical entry, the approaches share a certain notion of argument structure that basically predetermines the relation of the arguments to the verb. It is generally assumed that the lexical entry of the verb contains information on the number and type of arguments the verb will project into syntax. Certain approaches also distinguish between the external and internal arguments since it is argued that these nominals display distinct properties and occupy different positions within the verbal structure (Williams 1981, Zubizarreta 1982). In most views of argument structure, thematic roles represent the semantic properties of the arguments that are relevant to syntax, such as Agent or Patient, and encode the pertinent features of verb meanings. Others have argued that linking or projection of arguments should be formulated in terms of aspectual semantics rather than thematic relations (Ramchand 1997).

Certain linguistic phenomena raise a challenge for the fully predetermined argument structure approach. The "variable behavior verbs" raise an important problem for the lexical-entry based approaches since their valency or aspectual readings could vary depending on the syntactic context in which they appear (Borer 1994, Levin and Hovav 1986 for discussion).

example:

Examples from Dutch and Italian show that these verbs are alternately unaccusative or unergative based on the context they appear in. In these languages, the auxiliary have is used with so-called unaccusative and aspectually bounded verbs and the auxiliary be is used with unergatives and aspectually unbounded verbs. The choice of the auxiliary may vary, however, depending on other syntactic elements within the clause.

Variable behavior verbs have often been used to argue that membership in the unaccusative or unergative class is unstable and is not predetermined by the lexical entry of the verb. The causative alternations studied in the previous chapters are also problematic since, in these cases, the argument structure of a verbal element can be modified by the addition of an external argument or by changing the thematic roles and semantic properties of the arguments. In addition, the locative alternations exemplified below show that the arguments of a verb may be represented in different syntactic and thematic frames. Verbs that display multiple options for the expression of their arguments in syntax obviously cause a problem for the predetermined argument structure approach (see Ritter 1998 for discussion).

example:

Faced with the problem of alternating verbs, most lexicalist approaches opt for a computational lexicon with its own set of word-formation rules, which are able to derive the alternations on argument structure within the lexicon. Such a system can account for the close relation between the alternating verb pairs but leads to a proliferation of lexical entries since all alternations are stored as separate entries in the lexicon (cf. discussion in Alsina 1993).

Throughout this dissertation, I have argued for the elimination of the distinction between twoseparate components of verb-formation. The adoption of a single computational domain allows us to capture the close relation between argument structure and syntactic configuration without creating new prolific lexical entries and without positing two parallel components of verb-formation that end up duplicating each other's work.

In the proposed analysis, the arguments of the verb are not predetermined in the lexical entry but are interpreted compositionally based on the position that they occupy in the final structure. Hence, for instance, the arguments of a transitive verb are not marked as external or internal in the lexicon and in fact, are not listed within the verb's entry. Instead, they receive the properties of an external or internal argument depending on the structural position that they occupy. Thus, the projection of arguments is not predetermined but is construed within the syntactic structure by the combination of the various elements that form the verbal predicate.

In this chapter, I will investigate the relation between the arguments and the primitive elements of the syntactic code which combine to form the verbal predicate. The discussion focuses on the relation between the semantic interpretation of direct objects, the status of nominals in light verb constructions, and the structural position they occupy. It will be shown that properties such as specificity, aspectual interpretation and case-assignment can all be derived from the syntactic structure, without having to encode the information within a lexical entry.

Studies on various languages have pointed to the existence of two object types with distinct case morphology and different semantic readings. The data also show that the two object types occupy different positions in the phrase structure. In this chapter, I will show that Eastern Armenian also provides evidence for the existence of two distinct structural positions for the direct object, which correlate with case morphology and semantic interpretation. I argue, however, that languages should be distinguished depending on the nature of the semantic properties of the object that plays a role in case-assignment. In particular, I will show that in Eastern Armenian, Turkish, Persian and Hindi, the specificity properties of the direct object correspond to the type of case assigned. In contrast, case-marking in Finnish and Scottish Gaelic depends not on the specificity of the direct object but rather on its quantitative properties (i.e., whether the object NP represents a specified quantity). I propose an analysis that can account for the correspondence between object case and semantic interpretation in both language groups depending on the projection in which object case is realized. In this analysis, the inner verb phrase (or VP) is headed by a functional projection which I argue to be Aspect Phrase. This projection is closely tied to the quantitative properties of the direct object, since only objects expressing a specified quantity can delimit the aspect of the predicate. In addition, I posit an agreement projection AgroP outside the vP which has links with the specificity of the object NPs. Thus, case-assignment in languages such as Eastern Armenian is performed in the AgroP projection, wherease case-marking on objects in Finnish or Scottish Gaelic takes place in the Aspect Phrase. This analysis can capture the correlation between the type of case assigned to the direct object and the semantic interpretations obtained, be it specificity or aspect.

These findings suggest that the nominal properties of specificity and quantitativity should be represented in distinct projections in the syntax of the object NP. The analysis develops two parallel structures for the verbal and nominal domains, which are composed of corresponding primitive elements, where Number Phrase (NumP) carries the relevant information for the AspP projection, while DP properties correlate with an Agreement projection.

Following ideas developed in Vergnaud (2000) and Vergnaud and Zubizarreta (2001), Isuggest a theory in which the verbal predicate and nominal phrase each project their own domain in syntax. The two domains can enter into a checking relation at various points in the computation. In this proposal, the verbal component represents the temporal domain of language and the nominal structure denotes the physical domain; sentences are formed when the two parallel domains merge. Hence, the arguments of a verbal predicate are not predetermined in the lexicon and are not projected within the verbal structure but rather are members of a distinct parallel domain that enters into a checking relation with the verbal elements by merging into the verbal domain. Case-assignment can thus be treated as an overt realization of this checking relation.

The proposed architecture with two parallel verbal and nominal domains makes certain predictions with respect to the status of the nominal element in light verb constructions in Persian. It was argued in Chapter \ref{persian} that the preverbal element in these complex predicates is part of the verbal predicate and combines with categorial and functional elements in syntax to form the verb. These elements are formed from the root element of the verbal entry and provide the substantive information to the predicate. The current analysis, which distinguishes the verbal and nominal domains, also makes a clear distinction between the position occupied by the direct object argument and the preverbal element of a light verb construction. In particular, the proposed model states that the direct object belongs to the nominal domain whereas the preverbal nominal is part of the verbal domain.

Hence, in addition to the distinction between specific and non-specific objects in Persian and Eastern Armenian, I will also propose to differentiate the non-specific internal argument of a verb and the preverbal element of a light verb construction. The non-specific object is argued to be formed within the nominal domain and it becomes an argument of the verb when the two domains enter into a checking relation; the preverbal element, on the other hand, behaves as part of the verbal predicate and appears within the temporal or verbal structure. In other words, while the non-specific nominals are independent structures that merge with the verbal structure for checking purposes, the predicative nominals are members of the verbal structure itself and combine with the verbal functional elements to form a complex verbal predicate. This chapter thus distinguishes three degrees of argumenthood within the verb phrase in Eastern Armenian and Persian: a specific object, a non-specific object and a predicative nominal.

The formation of words has given rise to a lively debate in generative grammar which relates directly to the interaction between syntax and the word-formation component. The central question in this debate is whether the formation of words and the composition of phrasal predicates belong to distinct components of grammar or whether they are both derived within a single computational domain. This dissertation investigates the relation between morphology and syntax by focusing on the properties of linguistic expressions that lie at the word/phrase boundary. The close examination of complex verbal predicates and the projection of arguments led to the presentation of the framework of Parallel Domains, developed in Vergnaud and Zubizarreta (2001), and to the formalization of interface conditions.

This final chapter reiterates the questions that are at the core of this thesis and summarizes the framework that has been proposed in the preceding chapters. This chapter provides a picture of the emerging architecture and discusses some of the implications for current theories of grammar.

This chapter discusses the framework of "Parallel Domains" proposed, the notion of a lexical item, and phases as boundary conditions. The final chapter addresses some of the possible implications of the proposed framework for current models of computation and points to avenues for future research.