A central theme in this book's approach to lexical theory is that contextual features in individual entries must be formally simple, uniform in format, and stated in terms of psychologically accessible categories--i.e., categories which classify the concrete words of PF, not the abstract phrases of LF; as real as these latter are, they are not present in the data used by the child. That is, the many detailed properties of complex constructions (e.g., participles, nominalizations, passives) must derive from syntactically combining extremely transparent lexical entries. These entries can contain no diacritics or purely formal features.

The motivation for this is simply a classic Chomskyan argument from poverty of the stimulus. What individual children do in learning a specific language is acquire a hoard of lexical entries for "small" grammatical items, both bound and free. Now we observe that not only some but essentially every one of these learned grammatical items differs syntactically from its closest counterparts in closely related languages. For example, if there is a handful of grammatical free or bound English morphemes lexically specified as exact translations of French ones, they are exceptions which prove the rule (I have yet to encounter a single one). Given that pre-school or unschooled children learn hundreds of these language-particular grammatical items fast and well, their lexical representations must have a simple and uniform format, expressed in categories readily accessible in the linguistic data, i.e. subcategories of words.

On the other hand, the facts of rapid acquisition do not suggest that lexical insertion theory, presumably uniform across the species, has to itself be transparent, any more than transformational theory or phonological theory are transparent. The structure of the uniform lexical theory developed in this study is subject only to general considerations of parsimony, elegance and empirical coverage which we require of any scientific enterprise.

Curiously, most generative treatments of the lexicon, including those of Chomskyan inspiration, search for extreme simplicity just where it seems least likely to be found: areas of species-wide genetic predisposition which have taken eons to perfect are felt to be governed by simple statements such as "affect alpha." Yet in the area where poverty of the stimulus and rapid acquisition have some force, the largely non-genetic lexical lists, essentially any quasi-formalized linguistic or conceptual properties are attributed to individual items and no principles regulate either lexical form or the lexicon-syntax interface. I do not see how this combination of abstract principles operating on unconstrained lexicons contributes to explaining acquisition of actual languages. Worse, the unformalized nature of the lexicon renders largely untestable any empirical claims made for the derivational component of UG (which depends on "projection from the lexicon").

The literature under the rubric of Head Driven Phrase Structure Grammar, which human time limitations have prevented me from seriously investigating, does nonetheless seem to have a failing in this area as well, though one that is not so maddening as the hand-waving approach to the lexicon widely practiced in transformational syntax. While HPSG pays careful attention to formalizing the lexicon, it seems to care less about constraining it. Practitioners seem to be complacent and satisfied that if it works (i.e. can be computationally modeled), it's (relatively at least) good enough. In my tentative excursions into this literature, I sense little concern with trying to formulate a theory of lexical entries all and only of whose possible instantiations are realized in the lexicon of a language or some collection of languages.

This book therefore focuses on the urgent need for a formal, constrained and empirically revealing theory of a syntactic lexicon. To satisfy this need, it proposes new principles regulating subcategorization which determine how syntactic structures project from a language's lexicon. In the theory developed here, the trees projected by lexical subcategorization frames are not always copies of the frames themselves, as they are in the first classical proposals of Chomsky's (1965) Aspects of the Theory of Syntax. This study's highly restrictive sub-theories of grammatical categories and features, abstract case, derivational levels and economy principles have the consequence that sisterhood between a selecting and selected element is only subcategorization's "simplest case."

The data and paradigms here are for the most part drawn from English, but properties of Romance languages also form an essential part of the argumentation, and some points are discussed in terms of constructions from yet other languages, especially Japanese.

This study defends a strictly syntactic approach to the lexicon, i.e. it elaborates a theory of c-selection (subcategorization) and argues against the use of any thematic grids or lexical conceptual structures in grammatical computation. Constructions which have been widely invoked as necessarily involving semantic selection, such as the spray/ load alternation, propositional complements, light verbs and understood arguments are shown to be better analyzed without it (cf. especially Chapters 2, 6 and 9). A central organizing factor for this approach to the lexicon is a crucial distinction between an item's "cognitive syntactic" features F used in syntactic derivations and its "purely semantic" features ƒ which are not (Chapter 1). The first use of the former (Chapter 2) sharpens the theory of c-selection by using these features in lexical frames (e.g., +___ANIMATE rather than +___DP and +___PATH rather than +___PP).

The principal innovation based on the F/ƒ distinction is the proposal in Chapter 3 that the lexicon consists of two quite different components, a grammatical lexicon bereft of purely semantic features (the "Syntacticon") and a mental lexicon which consists of the open classes of the more specified contentful lexical items (the "Dictionary," which is the faculty of human linguistic memory and culture). There are only four categories in the Dictionary (N, V, A and P), what I term "nature's bottleneck" in a final discussion in Chapter 10. Dictionary items are always inserted at the outset of transformational computations on a domain, as in Chomsky (1965). The perennial question "why do transformations exist?" is answered as follows: to assemble sets of disparate open class items in structures which can be communicated at the Phonological Form interface (PF) and interpreted at the Logical Form interface (LF).

In contrast to the Dictionary, the Syntacticon is regulated by a theory of multi-level lexical insertion: the feature composition of a lexical item in the Syntacticon determines at which stage in a derivation it is inserted, i.e., satisfies its c-selection properties. I take it that sweeping statements about levels of lexical insertion are empty unless tied to predictions about particular items satisfying specific insertion contexts. Chapters 4 and 5 explore the predictive consequences of this study's theory of multi-level insertion for the Syntacticon's bound morphemes and Chapters 6 and 7 for its free morphemes. In the limiting case, entirely uninterpreted Syntacticon items are inserted at PF; in particular, the entire class of items traditionally called inflections are characterized as "late-inserted" in this way. Inflections which apparently contribute to meaning (past tense, noun plurals, etc.) in fact serve only to license empty nodes in other interpretable free morpheme positions (I, D).

While developing the theory of the Syntacticon and multi-level insertion, I argue not only against semantic selection but also against any autonomous "morphological component" with combinatorial properties, in the lexicon or elsewhere. I claim the only statements needed for morphology are those with phonological effects, possibly conditioned by syntactic factors. The entire content of morphology is thus akin to interface statements such as a rule of Classical Greek, "verbs and adjectives receive penultimate stress." Except for the effects of such statements, the combinatory principles of bound morphology are exactly the same as those governing the syntax of free morphemes, especially those of compounding (Chapter 3). Supposed differences between compounding and morphology in e.g. Romance languages are shown to be differences between free and bound morphemes.

The notion that categories in trees are associated with lexical items only in the course of derivations raises the question of how these categories act prior to such insertion. Many analyses in this book demonstrate that a category in a head position does not act like a head until it is lexically filled. This principle regulating underlying empty heads is introduced in Chapter 4 and used throughout the rest of the study. Since grammatical elements can be inserted at later derivational levels, the empirical properties of particular grammatical elements are typically due to (i) their delayed status as head of a construction and/ or (ii) whether they are interpreted in Logical Form. Simple and uniform lexical specifications leading to insertion of single morphemes at two or three derivational levels provide novel explanations for many previously poorly understood complex grammatical patterns, for example in both the passive and perfect periphrastic constructions of Germanic and Romance (Chapter 5).

In Chapter 6, Syntacticon entries of the form X, +___Y are shown to induce syntactic "flat structures" if both X and Y are the same category and X lacks purely semantic features ƒ (i.e., is subject to late lexical insertion). This peculiar conjunction of properties provides analyses which solve many recalcitrant syntactic puzzles of especially the Romance languages. Causative, restructuring, linking and light verbs, which have all been treated differently in the literature, actually realize very similar structures. The explanations all crucially exploit the notion of "empty underlying head." Another area which demonstrates the explanatory power of syntactically empty heads is a range of PPs headed by grammatical P, including adjuncts as well as complements (Chapter 7).

Chapter 8 builds to what is in some way the intellectual climax of the book. It begins by extending case theory and refining certain formal properties of subcategorization in a way that fully defines, in conjunction with the category and feature theory of Chapter 1, the notion of "possible syntactic part of a lexical entry." This chapter demonstrates that quite general and familiar classes of complements in the English lexicon instantiate all and only the structures predicted by the category and subcategorization theories of this book. This reformalization of the notion "lexical entry," after twenty years of promissory notes in the ever novel coinages of theta theory, semantic selection, conceptual structures and the like, finally proposes limits imposed by Universal Grammar on the construct "possible lexical entry." This step removes the vagueness that has been associated with the lexicon for decades and enables us to embed the study of syntactic derivations in a fully generative system, in which all modules are formalized.

The last two chapters pursue another hypothesis that privileges syntax over any active role for the lexicon in language use. These chapters argue that all understood arguments are represented syntactically; a case is made against various notions of "unprojected arguments" which have appeared in the literature. In particular, Chapter 9 justifies unexceptional, purely syntactic representations for "null complement anaphora" and "null generic objects," and derives obligatory PRO as a by-product of subcategorization for a V-headed rather than I- or C-headed complement. Finally, Chapter 10 argues for a discourse-governed reference of optional PRO subjects, which turn out to have a broader distribution than usually envisioned and include subjects of imperatives and agent phrases in verbal passives. All understood arguments are thus syntactically present arguments.

A potentially controversial psychological implication of this study is that language can express only those combinatory meanings which simply "arise" from valid syntactic combinations. In the view developed here, syntax determines entirely any propositions we can formulate about our mental world and yet is largely independent of and provides very little insight into how we otherwise conceptualize it. Moreover, a little reflection shows (see in particular the concluding sections of each of the last three chapters) that our mental world and the form of natural language are almost entirely incommensurable, at least in terms of our ability to consciously reflect on them. So for example, while models of propositions are always discrete and almost invariably two-dimensional, our conceptualizations are obviously continuous and three or four dimensional: The room slowly filled with smoke.

Thus, even if thematic relations between a verb and its arguments are just "convenient mnemonics for particularly prominent configurations" in conceptual structures (Jackendoff 1987: 385), a given verb's lexical conceptual structure, while psychologically applicable to the world, is still linguistically opaque and unlikely to correspond to any structure in a grammatical phrase marker outside the verb. Consequently, any pieces of conceptual structure attached to lexical items remain unanalyzable and are hence next to useless as a guide to more general knowledge about the interface of conceptual structure with syntax.