Chapter 3: Semi-lexical categories. nor truly functional. As Corver and van Riemsdijk rightly point out, There is more

Chapter 3: Semi-lexical categories 0 Introduction While lexical and functional categories are central to current approaches to syntax, it has been noticed that not all categories fit perfectly into this division. Some categories fall in between these two extremes: these categories are neither truly lexical nor truly functional. As Corver and van Riemsdijk rightly point out, There is more between functional words and content words than was hitherto assumed (2001: 17). The observation that there is something in between is captured in a new line of research that focuses on these semi-lexical categories. However, because this work is so new, there are a variety of approaches to the notion and there is no single definition of semilexical categories. In order to give the reader a flavor of how semi-lexical categories have been treated in the literature, I begin the chapter with an introduction to the concept and its possible definitions. In this introductory section, I will also briefly discuss how this idea has been used to treat categories such as nouns and verbs in order to show the range of word-types that can be covered by this framework and in order to illustrate the different approaches that have been forwarded under this rubric. After an introduction to semi-lexical categories, I turn to a discussion of how prepositions have been treated under the semi-lexical approach (Section 2), and in the following sections, I introduce the view of semi-lexical categories that will be adopted here (Section 3) and offer a preview of how this approach can be applied to the problem of categorizing adverbs, particles and two types of prepositions: semi-lexical and

47 functional. In the next chapter, the specific arguments in favor of the semi-lexical analysis offered here will be laid out. 1 Introduction to semi-lexical categories The concept of semi-lexical categories arises out of the observation that a strict bifurcation of lexical and functional categories cannot account for all elements. Corver and van Riemsdijk (2001), in their outline of the development of the notion of semilexical categories, point to work by Ross that argued against such a strict bifurcation in the early 1970s. Ross (1972, 1973) already argued that the traditional view of syntactic categories as being discrete elements that can be rigidly distinguished from one another is incorrect. Instead of a fixed, discrete inventory of syntactic categories, he proposes a quasi-continuum, where the distinction between one category and another one is not discrete, but squishy, i.e. one of degree. According to this view, syntactic categories are not, or not necessarily, distinguished from each other by the absolute presence versus absence of some property X. Categories rather differ from each other in relative terms: that is category a (say V) may have property X to the highest degree, category b (say A) to a lesser degree, while for category g (say N), this property is nearly absent or maybe even completely so. (Corver and van Riemsdijk 2001: 4) The question of gradience 1 can be applied not only to lexical categories, but also to the dichotomy between lexical and functional categories. Emonds (1985) points to a set of words that falls in between the lexical and functional categories. These words are subclasses of the lexical categories, and are taken to be a grammatical instantiation of a lexical element (Emonds calls them grammatical nouns, etc.). The two most prominent 1 The idea of gradience is introduced only to point out that there may be a wider distribution of categories than is currently thought. The approach adopted here will aim to describe a discrete, systematic view of categories.

48 characteristics of these elements are their high frequency of usage and a lack of substantive content. Representative examples are given in (1). (1) a Grammatical nouns: one, self, thing, place, body b Grammatical verbs: be, have, get, do, make, let, say c Grammatical adjectives: other, same, different, such d Grammatical prepositions: out, up (i.e. particles) Another example can be found in van Riemsdijk s (1998) examination of the Direct Partitive Constructions in Dutch and German; he is led to the conclusion that the first noun in these constructions is a borderline case: it is neither a truly lexical head (i.c. [sic] a lexical N), nor a truly functional head (Corver and van Riemsdijk 2001: 9). Because this element has characteristics of both lexical and functional elements, he used the term semi-lexical to describe this in-between state. Current work on semi-lexical categories offers no single, unified definition of semi-lexicality; various researchers offer a variety of different definitions following one of three general themes. First, there is the view that semi-lexical heads are lexical heads that lack semantic content. Emonds (2001), drawing from his earlier work (1985), states that semi-lexical heads are those N, V, A, and P which have no purely semantic features (people, thing, do, get, much, so, by, of, etc.) (p. 29). Schütze (2001), following Emonds approach, defines semi-lexical heads as fully lexical heads that lack intrinsic semantic content (p. 127). Second, there is the view that semi-lexical heads differ from functional heads in their semantic features (this differs from Emonds and Schütze, where semi-lexical elements lack content; in this view they have content, and this is what distinguishes them), but not their functional ones. Powers (2001) work on child

49 language acquisition exemplifies this view: semi-lexical heads are closed-class elements that are distinguishable based on their semantic, not their syntactic, features. Drawing on the earlier work of van Riemsdijk (1998) on the status of categories lying between the prototypical lexical and functional categories, the third view takes semi-lexical heads to be a mix of lexical and functional properties. Bhattacharya (2001) states that a semilexical head exhibits several properties which are divided between lexical and functional heads (p. 191). The ways in which these definitions are instantiated in the syntax also differ. Zeller (2001), for instance, sees semi-lexical heads as consisting of two morphemes: a lexical head and a functional suffix. Cardinaletti and Giusti (2001) define semi-lexical heads as lexical categories merged as functional heads (p. 372). And Haider (2001) argues for two types of semi-lexical heads: one is a lexical head without A-selection (i.e. a lexical head with functional qualities) and the other is a functional head that has lexical content (a functional head with lexical qualities). Even the status of semi-lexical categories as unique categories is unclear. Most researchers do not explicitly address whether or not semi-lexical categories comprise a third, distinct category in addition to the lexical and functional ones. Instead, they leave open the status of the category. Except for Cardinaletti and Giusti (2001), who unequivocally reject the idea that semi-lexical categories should be considered as a third category, most seem to imply either an autonomous category or some sort of dependent category. Butt and Geuder (2001), for example, in their examination of the semantic and syntactic patterns of light verbs, conclude that not only are light verbs a semi-lexical

50 category, but they have to be considered as a class in their own right (p. 323). Haider (2001), on the other hand, does not view the semi-lexical category as an independent category, but as a term applied to all elements that are outside of the canonical lexical or functional categories: Semi-lexical heads [do] not constitute a discrete category but rather a fuzzy set of elements in between the two extremes, namely a pure structural head, and a fully furnished lexical head (p. 68). Despite the lack of agreement on the definition and the status of semi-lexical elements as categories, quite a variety of elements have been studied under the rubric of semi-lexical categories. Classifiers, quantifiers, motion verbs, serial verbs, and light verbs have been the focus of recent studies. I briefly outline the main themes in several representative papers in order to present the reader with a flavor of the types of structures that have been examined in this field. Bhattacharya (2001) argues that the Bengali complex head consisting of a quantifier or numeral used in conjunction with a classifier (the Numeral/Quantifier- Classifier complex) is semi-lexical because it exhibits both lexical features (for example, quantifier floating is allowed, thus separating the head from its complement) and functional ones (such as the lack of descriptive content and the restrictive, closed-class nature of the class). Veselovská (2001) argues that there are three types of quantifiers in Czech, based on agreement, ellipsis and relativization patterns. She argues for a semi-lexical interpretation of the class of quantifiers due to their restricted semantic content, in contrast to lexical nouns and adjectives, which have a full semantic content.

51 Light verbs are argued to be semi-lexical in the work of Butt and Geuder (2001) in Urdu and Wagiman (both having V+V constructions), and English (having V+N constructions); light verbs are differentiated from auxiliaries on syntactic and lexical grounds. Syntactically, light verbs pattern in some ways like a main verb (most importantly, it carries the inflectional information that would normally be carried on a main verb) and in some ways differently from the main verb (the light verb cannot be topicalized without the verb, while the main verb can topicalize away from the light verb). They also cannot be taken as auxiliaries, due to differences in syntactic patterning in word order, case marking, topicalization, and restrictions on tense or aspect. Lexically, they are seen as having a minimal semantic contribution: they are defective verbs in that they make no independent reference to a class of events. They contribute additional semantic features to the event description (p. 358). However, the semantic contribution of the light verb is distinguished from aspect, in that the auxiliary provides aspect (it situates the event), while the light verb modifies the event, by contributing, for example, a path reading to the event. This places light verbs in the middle of a continuum with lexical verbs and auxiliaries forming the lexical and functional limits. Interestingly, light verbs in Urdu all have fully lexical counterparts, which is seen as a case of polysemy because light verb constructions can be understood as continual extensions from the prototypical concrete meaning of [the lexical verb], with the extension of the meaning being correlated with the increasing abstractness of the referents filling the THEME argument slot (p. 343). Butt and Geuder refine this argument by broadening it to account for both V+V and N+V constructions found cross-

52 linguistically; essentially, they conclude that light verbs play a modifier-like role in the verbal domain: they modulate the contents of the semantic roles of the full verb s arguments (p. 364). In examining the serial verb constructions of the Creole São Tomense, Hagemeijer (2001) argues that the first position verb (V1) and second position verb (V2) create a complex predicate (two argument structures acting as one), and that V1 and V2 are defective in different ways. V1 is complete in all of its functional aspects as a verb, as it can take Inflection, assign Case, and assign a theta-role (if transitive), but it is lexically defective because while it alters the semantic content of V2, it must rely on V2 for its argument structure. V1s also comprise a restricted, or semi-open class of verbs; not just any verb can be used in V1. In contrast, V2s are defective because they cannot carry Inflection or negation (they can assign Case and theta-roles, but only when they are reanalyzed as prepositions). Hagemeijer argues that V2 are categorially ambiguous, as they waiver between having verbal and prepositional features. V2s can be seen as existing on a continuum that ranges from prepositional to hybrid status (verbalprepositional) to a state closer to full-fledged verbs (p. 445). Hagemeijer concludes with a brief discussion of the possible directions of grammaticalization that may occur with semi-lexical heads: They may fully functionalize, which is arguably what happened in many cases to TMA markers, which in Creoles are often claimed to derive from verbs; they may receive full specifications again, i.e. lexicalize, when V2 becomes, for example, relabeled as a preposition; finally, they may remain semantically underspecified up to different degrees, like Mandarin s ba, Fon s só, etc. In the light of the latter scenario, a tendency toward full functionalization may be expected over time. (p. 445)

53 While most of the work in semi-lexical categories implicitly or explicitly argues for the existence of some sort of semi-lexical category, Cardinaletti and Giusti (2001) argue against such an analysis. Their work focuses on motion verbs that occur in inflected constructions, such as I go buy bread; this construction is examined in Romance and Germanic languages such as Italian, English and Swedish. They argue that there are no constant features that allow motion verbs to be consistently identified as a single category across all languages. However, while the specific features of motion verbs differ cross-linguistically, it is true that they always exhibit a hybrid behavior that combines functional features (closed class, placement parallel to auxiliaries, etc.) and lexical ones (semantic content, selection of a subordinative-like element, etc.). In light of this, they argue in favor of a definition of semi-lexicality that combines the (various) functional and lexical features of motion verbs, but doesn t add a third (semi-lexical) category. Their proposal is that motion verbs are actually lexical heads that lack some of their typical lexical properties due to their merging into a non-lexical (functional) projection (p. 407), and they claim that this definition of semi-lexical elements as lexical categories merged as functional heads captures the fact that all the functional properties that can be claimed for motion verbs in the cases under consideration are actually lack or suppression of their canonical lexical properties (p. 372). Clearly, there are a number of unresolved issues regarding semi-lexical elements. First, there is no consensus regarding the definition of the notion semi-lexical, and there is disagreement with regard to the status of semi-lexical elements as a category.

54 However, there is one clear, unifying theme that is identifiable in each of the various articles on semi-lexical elements: there are items that do not neatly fit into the lexicalfunctional division of categories. Almost always, these elements are difficult to categorize because they display a mix of functional and lexical features, or are defective in lacking some proto-typical feature that they would have if they were purely functional or purely lexical. 2 Prepositions as semi-lexical categories In addition to the work on semi-lexical elements in the nominal and verbal domain, there has been some work done on the semi-lexical status of prepositional elements. There are two main constructions that have been the focus of study. The first examines the use of prepositional complementizers in small clause constructions; the second concentrates on the complicated situation found in German where there are a variety of adpositions, including prepositions, postpositions, circumpositions, and particles. As with the work on semi-lexical categories in other domains, the work on prepositions or preposition-like elements varies widely in how semi-lexicality is defined and how the constructions are used to argue theoretical angles. Eide and Afarli s (2001) central claim is that semantics must be taken as central to syntax. The basis of their work relies on Bowers (1993) notion of the predication operator, which essentially acts to link the subject and the predicate in any given clause. Importantly, this operator has its own functional projection, but it can be lexicalized in that it can be filled with lexical information. Focusing on Norwegian structures similar to English phrases like I regard [John as crazy], they argue that in the small clause, the

55 object of as becomes a predicate due to the predication operator found in the small clause (following Bowers 1993). The operator is taken to contribute predicative content, while the lexicalized element as (in this case) acts primarily as a structural marker. Possibly it adds a hypothetical sense to the clause. They assume a dynamic Principle of Compositionality (Frege s Principle), where the meaning of the whole phrase consists of the meaning of the parts plus the meaning that is created in combining the parts. Because the operator can be filled with a variety of elements, and because these visible elements can occur in different operator positions, it is argued that the syntactico-semantic substance resides, not primarily in the supporting element, but in the abstract operator made visible by the element (p. 461). Given this, they argue that functional projections are all projections of operators, and that it is the basic operator structure that gives the sentence its basic, underlying logical form. In essence, all functional projections are semi-lexical, in that they are really operators that are (generally) filled by lexical material. Like Eide and Afarli, Rafel (2001) focuses on the particles (to use Rafel s terminology) as and for in phrases such as They regard John as smart and They took him for a fool. But she has a different goal in mind: her focus is the structure of the small clause itself. She argues that the standard approach, Stowell s Small Clause Theory (1981, 1983), must be replaced. Under the standard analysis, given in (2), the DP John is base-generated in the small clause, is assigned a theta role from the predicate (an AP in (2a) and a DP in (2b)), and moves for Case assignment. The particles as and for are taken to be prepositional complementizers.

g 56 (2) a They regard John i [CP t i [C as... [Scl (XP) t i smart]]] b They took John i [CP t i [C for... [Scl (XP) t i a fool]]] Rafel points to the following problems in the structure: 1) the prepositional complementizer for doesn t assign case to the DP John; 2) the prepositional complementizer licenses a trace, when it usually doesn t; 3) movement of the DP creates a mixed chain ([A...A...A]), so the structure should be ungrammatical; 4) in languages like Spanish, where Case can generally be checked at Logical Form (LF), the DP can never remain in its base-generated position (i.e. Case cannot be checked at LF here); 5) verbs like take don t subcategorize for a full CP, but only a small clause. Given these problems, Rafel offers an analysis that parallels a CP-IP construction. As seen in (3), Rafel s Complex Small Clause analysis relies on the use of PRO to overcome the problems of the standard analysis. (3) CPx 5 Spec Cx Johni 4 Johni Cx FPx g as for FPx gp FX XP 2 Spec X PROi PROi X smart fool Based on the observations that the particles occupy the highest extended projection of the lexical head (A or D) and that they contribute to the semantics of the matrix verb, Rafel concludes that the particle...behaves like a semi-lexical head in the sense that is

57 functions as both the highest functional head that introduces the [Small Clause]-domain and the lexical elements that heads the [Complex Small Clause] (p. 488). While the work on prepositional complementizers in small clauses focuses on a very narrow set of preposition-like elements, there is at least one paper that examines a wider range of elements. In trying to sort out the complex picture of adpositions in German and Dutch, Zeller (2001) attempts to account for much larger sets of prepositions. Zeller s (2001) work defines semi-lexical elements in morphological terms: an element is considered to be semi-lexical when it is comprised of a lexical node and a functional suffix. In centering his attention on German and Dutch prepositions, postpositions and particles, he argues that postpositions are semi-lexical. Postpositions occupy functional heads, but are not truly functional (like tense or Det) because they are derived from lexical prepositions. 2 Essentially, the postposition is a preposition with a Zero-operator, a suffix that is functional and changes the thematic properties of the postposition. Thus the postposition is a complex head, where the semantic content is contributed by the lexical preposition and the categorial features come from the functional zero-suffix. Lastly, Zeller uses the idea of environment to distinguish between the different adpositional elements found in German. Under this model, prepositions, particles and postpositions can all be distinguished depending on their syntactic context. Prepositions occur when P is adjacent to a functional head; particles 2 This raises interesting questions for the infinitival to in English.

58 occur when the P is adjacent to a verb; postpositions occur when the P has a functional suffix. It is important to note that there have been no attempts to apply the notion of semi-lexical categories to the whole class of English prepositions. To date, the only study that has tried to account for a wide range of (German) prepositions has been Zeller (2001); the remaining studies have focused on the very narrow domain of prepositional complementizers in small clause constructions. It is the goal of this thesis to present a preliminary framework that uses the notion of semi-lexical categories to account for the two pervasive problems found in work on English prepositions. First, the idea of semi-lexical categories can explain the relationship between prepositions, adverbs and particles in a way that captures the observations that have given rise to two competing views of these elements. I will argue that the reason that these three categories seem simultaneously different and the same is because they are all instantiations of different elements that belong to one domain. In pursuit of this central theme, I also attempt to show that prepositions proper must be divided along the lines indicated by current research: there must be some prepositions that are functional and others that are not. The remainder of this chapter is devoted to spelling out the details of this program. 3 Defining a semi-lexical approach to categories The most general view of semi-lexical categories is that they lie somewhere between the lexical and functional categories--that they contain some lexical features as well as some functional features. Thus, semi-lexical categories, then, are generally taken

59 to fill the middle ground between these two fundamental extremes. This view can be visualized in a diagram such as the one in Figure (3.1). Lexical Semi- Functional lexical Figure (3.1): Three categories: lexical, functional, and semi-lexical In the nominal domain, for example, nouns such as dog and cat would anchor the lexical side of the spectrum, while determiners like the and a would anchor the other. In between would be the semi-lexical elements like thing, one, and body, as discussed in Emonds (1985). The same picture would emerge for verbs, where verbs anchor one end of the spectrum, and INFL (or more specifically, Tense, Aspect, Agreement, etc.) anchors the functional end. In between will be elements such as be, have, and get. Fundamentally, this approach accurately captures the fact that there is something that lies between the functional and lexical extremes. However, the situation is a little more complex than the one illustrated in figure (1). While semi-lexical categories are combinations of lexical and functional features, there must also be a place for elements that are essentially the opposite of semi-lexical categories. If semi-lexical categories are conglomerations of lexical and functional features, then there are also probably elements that are related to the category but lack lexical and functional features. This category is indicated by the question marks in Figure (3.2). Lexical Semi- Functional lexical???

60 Figure (3.2): Four categories: lexical, functional, semi-lexical, and??? These four categories, lexical, functional, semi-lexical, and???, comprise the four basic divisions that make up any given category. Nouns and Determiners, for example, would be taken as two basic categories that make up part of a larger umbrella category, as would Verbs and Inflection. In one sense, we already have this view of categories: when we want to focus on the fact that these two categories are inherently and intimately linked, we say that they are both part of the nominal domain or the verbal; when we want to focus on their differences, we focus on them as Nouns (or NPs) and Determiners (or DPs). Grimshaw (1991) has already pointed to the idea that the lexical head and functional head of the same projection must belong to the same category, so that verbs and Infl are both [+V, N], but the difference between the two is that verbs are non-functional while Infl is functional. This view needs to be expanded in light of the evidence that semi-lexical categories exist. It needs to include a wider range of possible categories: semi-lexical elements and their logically possible opposites. And due to the inherently close relationship between the two base categories (N and D, or V and Infl in these examples), we should focus less completely on the differences between these categories and recognize their functions as individual domains. In the remaining discussion and chapters, I will use the term domain to refer to the overall (umbrella) category, and category to refer to the division of the domain. I will argue that the domain is comprised of four categories. Each of these four categories is different from the others, and in that way they can be seen as opposites of one another: lexical items will be the opposite of functional ones, and the semi-lexical

61 category will be the opposite of an (as yet) un-named category. Thus, while each category differs in its syntactic and semantic characteristics, they are all members of one family: the prepositional domain. This view underscores their differences as well as their similarities, and allows us to see them as part of a complex category. In the next section, I will more fully and explicitly define how the domain can be subdivided. I then turn to a preview of how these divisions can be applied to the prepositional domain (which I expand in a detailed analysis of each prepositional element in the next chapter). I will argue throughout that viewing traditional syntactic categories as domains offers a logical way to account for the problematic nature of prepositions: it resolves the problem of the relationship between adverbs, particles and prepositions, and it incorporates the observation that there must be more than one type of preposition in the functional-lexical divide. Of course, it is important to keep in mind that while my discussion will focus solely on prepositions, the notion of this spectrum should apply to all of the major categories; I leave this to future research. 3.1 Defining the domain First, and central to the hypothesis laid out here, any element can have a combination of lexical or functional information. In order to provide an initial working model, I will adopt a binary features approach 3 parallel to the one suggested by Chomsky (1970) for distinguishing the four major lexical categories (or, domains in the terminology used here), only these features will define the categories that make up each

62 of the four major domains: Noun, Verb, Adjective, and Preposition. The two features proposed mirror Grimshaw (1991) and van Riemsdijk s (1991, 1998) [±F] feature which is used to divide the major lexical categories (domains) into functional [+F] and lexical [ L] elements. The central difference is that a more fine-grained approach is used: each category within a domain will contain a mixture of lexical and functional information (rather than a simple opposition of functional versus lexical information), represented in terms of positive and negative features that are assigned to each element. The two features that will be used are [±Lexical] and [±Functional]. 4 As described below, the Lexical features will refer to the availability of notional, descriptive content, and the Functional features will refer to the element s syntactic contribution. 3.1.1 [±Functional] features At their heart, functional elements are fundamentally those that provide connectivity in a phrase or sentence. The role of INFL in a sentence, in essence, is to link the subject and the predicate phrases into a single, interpretable phrase. Similarly, the role of the CP can be taken as the link between the discourse-level information and the resulting structure; it ensures that the appropriate transformations or movements occur if a question is the targeted pragmatic structure. In domains that allow Case assignment (verbs and prepositions), the linking function that is performed by these 3 I acknowledge the fact that a binary features approach must at some level inherently simplify a complex situation; I hope that future iterations of this hypothesis may be able to alleviate this intrinsic problem by finding a better solution. 4 I adopt the terms functional and lexical because (1) I wish to keep the continuity of what has been suggested before and build on it, and (2) because I believe that this distinction has something interesting to offer theories of categorization, but that it needs to be refined in order to be useful.

63 elements follows the Case assignment. In the verbal domain, for example, the VP semantically selects for the arguments of the predicate, but it is the inflectional features of the IP that license the subject. Thus, while the semantic structure of the predicate is determined in the VP, it is the IP that (syntactically) links the subject and predicate through the assignment of Nominative Case (without Case the subject will not be licensed). In some instances, when Nominative Case is unavailable, an Exceptional Case Marking verb can license the subject of an embedded clause by assigning Accusative Case. In these structures, it is the verb s Case assigning properties that contribute the linking features. Similarly, the Case assigning properties of transitive verbs license the verbal complement, and thus links the verb to the phrase that constitutes its object. If the primary purpose of the IP is to license the relationship between the subject and the verb, then Agreement and Tense should also be taken as examples of linking properties. Agreement essentially ensures that the elements in different domains (for example, determiners, modifiers and nouns in the nominal domain) carry similar features (person, number, gender, etc.), and in effect links them together. Tense, also, links elements; it ensures that the proper temporal features are found in all segments of the verb phrase. In the verbal domain, then, we see that Inflection and transitive verbs, as shown by Case assignment and Agreement, have this linking quality that is instrumental to being functional. The prepositional domain mirrors the verbal domain in its ability to assign Case, and in some languages, prepositions can also carry agreement features (as in Welsh): thus prepositional domain also has the possibility to link elements. Within the nominal domain, Agreement features ensure that all elements within the DP are properly

64 aligned with regard to things like number and gender. And while nouns themselves do not assign Case, there is (Genitive) Case assignment in this domain, as well. In conclusion, the most salient quality of the Functional feature is its ability to tie or link elements in a phrase together. This linking ability generally manifests itself through Case assignment or Agreement features. Thus elements that can assign Case or Agreement are designated [+Functional], while those that can t are designated [ Functional]. 3.1.2 [±Lexical] features While the Functional feature is used to capture the syntactic side of the story, the Lexical feature focuses on the semantic side. The [+Lexical] designation is tied to notional content: elements that are [+Lexical] are those that contribute a notional, or substantive, content. They form the substantive, semantic foundation of the sentence by denoting actions (Verbs), referents (Nouns), properties (Adjectives), and relationships (some types of Prepositions). On the other hand, elements with a [ Lexical] designation may contribute abstract information (tense, aspect, number, and so on), but offer no notional content. Elements such as Infl, D, and C are taken as the prototypical examples of the [ Lexical] designation. One characteristic of descriptive, [+Lexical] elements is that they contribute meaning compositionally: each element contributes it s own content independent of other elements. Because of this compositionality, [+Lexical] elements can easily be

65 substituted to create different meanings. This is illustrated for nouns (4a), adjectives (4b), and verbs (4c). (4) a The girl/dog/kitten/butterfly is cute. b That dog is cute/brown/small/fuzzy. c The boy kicked/hugged/bathed/fed his baby brother. Elements that contribute a non-descriptive, abstract level of meaning seem to do so in a non-compositional way that relies on some other element. For example, expression of number must be tied a noun, comparative and superlative markers must be tied to adjectives, and inflection must occur with verbs. These elements are not allowed to appear isolated from their descriptive counterparts (even in citation forms, we spell out the morphemes, rather than pronounce them as whole units), and these elements cannot be substituted for one another. As seen in (5), plural markers, superlative markers, and tense markers cannot be substituted for one another: they must occur in certain, set combinations. (5) a men / *mans / *manen (as in oxen) / mani (as in cacti) b more beautiful / *beautifuller c ate / *eated A second characteristic that may be valuable in distinguishing the two levels of Lexical features is whether or not the element can play a role in the argument structure of a sentence. Because arguments are the substantive elements that are required by a predicate, [ Lexical] elements, by virtue of their abstract type of meaning, should not be able to act as an argument in a given structure. [+Lexical] elements on the other hand should be able to. Similarly, while adjuncts are not required by the predicate, they do act to modify the predicate in some substantial way. In fact, by definition, adjuncts are

66 components which provide information about time, manner, reason, place, modality, and so on, of the event or state of affairs expressed in the sentence (Haegeman & Guéron, 1999: 29). Thus [+Lexical] elements would also be expected to be able to act as adjuncts, while [ Lexical] elements would not be able to. In conclusion, [±Lexical] features are tied to the degree of semantic contribution. [+Lexical] elements contribute a descriptive, substantive meaning, while [ Lexical] elements don t. Because of the differences in the type of semantic contribution, [+Lexical] elements contribute non-compositionally, and can act as arguments or adjuncts. [ Lexical] elements, on the other hand, contribute abstract meaning compositionally and cannot act as arguments or adjuncts. 3.1.3 Four categories in each domain In traditional discussions of categories, only two distinctions are made: categories are either lexical or functional. In the approach adopted here, we can offer a more explicit and precise view of these two broad categorial distinctions. The binary features for each of these purely functional and purely lexical elements are shown in (6) and (7): (6) Functional head: [ Lexical, +Functional] (7) Lexical head: [+Lexical, Functional] This designation of the binary features explicitly captures the generalizations that intended by the opposition of lexical and functional features in the general syntactic literature. Not all elements that have traditionally been described as lexical will carry this particular feature bundle; purely lexical elements are [+Lexical, Functional], but other

67 lexical elements are [+Lexical, Functional]. Much of the attention given to understanding syntactic categories has focused on discovering and defining the functional categories, leaving lexical categories to remain much as they have in the past (minus a few functional elements). When more attention is paid to divergent patterns found in the lexical categories, we should find that not all lexical elements are purely lexical, and some may have more functional content than was hitherto assumed. Certainly, I will argue this point for the category of prepositions. 5 As we have seen in the literature on semi-lexical categories, semi-lexical elements are generally taken to represent a combination of lexical and functional information. I will adopt this view of semi-lexical elements. Because these elements simultaneously play a lexical and a functional role, they are assigned a positive feature for both the Lexical and Functional features. The feature matrix of semi-lexical elements is as in (8): (8) Semi-lexical head: [+Lexical, +Functional] The remaining feature matrix consists of negative values for each feature. These are items that lack both lexical and functional content, and thus rely on another element of the sentence for their full interpretation. While they do impart meaning to the phrase, it is only in conjunction with another element, and the meaning contributed by the element itself will not be referential or substantive. On the functional level, these elements also lack the key features that would define them as functional: they cannot 5 As noted earlier, this approach should have wide-spread application in other domains as well. I leave this to future research.

68 assign Case. These elements can loosely be termed idiomatic. This feature matrix is shown in (9): (9) Idiomatic head: [ Lexical, Functional] In conclusion, this view of categories gives us four logical possibilities for categories, two of which cover the widely accepted, vital lexical and functional categories. The two additional categories consist of a semi-lexical category that contains both lexical and functional information, and an idiomatic element that carries no functional and no lexical information. It must be kept in mind that these four categories do not comprise monolithic categories filled with words that are identical to one another. Just as transitivity can be used to discuss types of verbs, there are individual differences found within each type. The same is true of these categories: while every semi-lexical element will carry the features [+Lexical, +Functional], each individual semi-lexical element will have its own unique properties in addition to these features: properties of meaning, patterns of interaction with other elements, etc. These [±Lexical] and [±Functional] do not determine every detail of the individual elements, but specify a very broad level of syntactic and semantic interaction. 6 3.2 Speculations on the structure of the domain Before turning to a preview of how prepositions can be taken as an example of a category that displays this full range of options, I offer a few speculative remarks on the possible structure and nature of the features themselves. I hope that further

69 investigation into a broader set of domains (verbs, nouns, adjectives, etc.) and deeper insights into how individual words interact with one another in each domain will strengthen and clarify these points. Following mainstream accounts of the relationship between lexical and functional heads, I assume that functional projections dominate lexical ones. The only level of information that must be added are the [±Functional] and [±Lexical] features that lead to the fine-grained approach presented above; this information is posited in the Spec(ifier) position, similar to Wh-features in the [Spec, CP]. In (8), the functional projection (FP) dominates the lexical projection (LP); in the [Spec, FP] are the [±Functional] features, and in the [Spec, LP] are the [±Lexical] features. 7 (10) FP 3 [±F] F 3 F LP 3 [±L] L 3 L One might also wonder where the features are found in the grammatical system: are they properties of structures or of the lexical items that go into the structure? It seems that the features [±Lexical] and [±Functional] must be found conjointly in the lexical entries (i.e. in the subcategorization frames) and in the structure itself. As seen 6 The broadest level would be the one that determines the domain of word: [±N, ±V] (if these two values are indeed the correct features to capture the differences between the syntactic domains). The [±Lexical, Functional] features define each element s use/relationship within each of these domains. 7 Given limitations on time and space, I leave the details of how these projections interact to further investigation.

70 above, these features are inherent to the projection of structure. However there the feature must also be found at the word level, as it is clear that not just any lexeme can act as a functional element just by virtue of its placement in a functional projection; the lexeme that is placed in a functional projection must carry (or have the possibility of carrying) [+Functional] features at a more basic level (in its lexical entry). Similarly, a purely lexical element cannot fulfill the requirements of a purely functional projection. Thus each lexical entry contains the specification of features, which then map onto the appropriate structure where it appears. Probably, there is an underspecification of features, so as to allow for flexibility in the system; this would account for the widespread homophony that we find in syntactic domains like the prepositional domain, where many lexical elements (carrying the features [+Lexical, Functional]) can also be used as semi-lexical elements ([+Lexical, Functional]), but are never used as purely functional elements ([ Lexical, +Functional]). Even the very restricted membership of the set of purely functional elements can overlap with other elements; as will be apparent in Chapter 4, we find overlap in the prepositional domain between the functional elements and the semi-lexical elements. But there cannot be a complete lack of feature specification, otherwise it would be impossible to account for the restrictions that do occur. For example, in the prepositional domain, purely functional elements never seem to overlap with purely lexical ones. Again, I must leave working out the details to future investigation.

71 3.3 The prepositional domain: a preview We have observed that the status of adverbs, particles and prepositions is unclear. The syntactic distribution of these elements shows that they are different from each other, but their nearly complete overlap in phonological form and their semantic similarity indicates a closeness of relation. Adverbs, particles and prepositions are in fact part of the same domain, but are different in that they represent four different categories. Adverbs, because they contribute descriptive content to the semantics of a verb phrase, but add little or nothing to the syntax, represent the purely lexical element in the domain. On the purely functional side are prepositions, like of, which contribute solely to the syntax, and contribute no substantive content. These functional prepositions contrast with traditional prepositions like in, on, under, with, and for, which contribute both substantive information and grammatical information (through case assignment). These represent the semi-lexical area of the domain, and will be called semi-lexical prepositions. Particles represent the final segment of the prepositional domain. The particle in and of itself doesn t contribute substantive meaning (in fact, den Dikken calls particles nonlexical elements (1995: 270)), and they don t contribute to the grammar of the phrase either. Particles represent the idiomatic level of the category. Altogether, the four categories of the prepositional domain have the following feature matrices (11): (11) Adverb: [+Lexical, Functional] Particle: [ Lexical, Functional] Semi-lexical Preposition: [+Lexical, +Functional] Functional Preposition: [ Lexical, +Functional]

72 There are several immediate advantages to this approach. First, it solves the problem of how adverbs, particles and prepositions are related to each other: they are members of the same domain (perhaps [ N, V] under Chomsky s (1970) features), but they are also different from one another in their behavior and their contributions to the meaning and grammar of the sentence. It also allows the lexical and functional nature of prepositions, which has been under discussion in the literature, to be clearly seen. Lastly, as we will see when we turn to first language acquisition for evidence of these differences, this approach makes a clear prediction about language acquisition patterns, and gives us a clear method to test the validity of this approach. In the next chapter, I will more fully discuss the lexical and functional qualities of each of the four types of preposition in turn.