Logical semanticists have standardly analyzed definite and indefinite descriptions as quantifiers, and definite pronouns as variables. This dissertation explores an alternative analysis, according to which all definites and indefinites are quantifier-free, i.e., consist of an essential free variable and the descriptive predicate (if any). For instance, "A cat arrived" is analyzed as "cat(x) & arrived(x)." The existential force that such sentences carry in their unembedded (and some of their embedded) uses is attributed not to the indefinite article, but to principles that govern the interpretation of variables in general. The primary motivation for this variable-analysis is that is leads to a straightforward account of indefinites serving as antecedents for pronouns outside their scope, as in the so-called "donkey sentences." These are handled by combining the variable-analysis of indefinites with a treatment of quantifiers as basically unselective, drawing on work by David Lewis. Since a variable-analysis of both definites and indefinites prima facie obliterates the distinctions between the two, it must be accompanied by a new theory of the definite-indefinite contrast. This theory must account for the different conditions under which definites and indefinites can get bound, and for the exclusive capacity of definites for deixis and anaphora. All these differences can be predicted if the uniform semantic analysis of definites and indefinites is supplemented by suitable assumptions about their contrasting felicity conditions (presuppositions): Felicitous definites must be "familiar" variables, felicitous indefinites must be "novel" variables. Familiarity may be a matter of having an antecedent in the text, or else of having a contextually salient referent. The underlying concept of familiarity that unites these two cases is best captured in an enriched semantic theory, which includes a further level of analysis, here called the "file" level. A major part of the dissertation is devoted to developing and motivating such a theory of "file change semantics," and to making precise its relation to conventional truth-conditional semantics.

This thesis investigates the system governing complex morphological forms in English and the way in which this system is acquired. Theories to account for constraints on the complementation of derived forms have been developed by Carlson and Roeper (1980) and Roeper (1981b). In elaborating and critiquing these theories, we are led to reconsider morphologically complex items in terms of the operations by which they are derived, and the "lexical level" at which each of these operations takes place. In addition, a principle of morphological "Inheritance" is proposed. This principle determines when a derived form may inherit the complement structure of its base, given a defined measure of morphological "distance". Together, the theory of lexical levels and the Inheritance Principle account for the restrictions on complementation in derived forms. The question of how formal linguistic systems are acquired is one which theories of language acquisition address; the second part of this thesis examines the morphological proposals in this context. We formulate a learning model which incorporates the Inheritance Principle and a particular theory of Universal Grammar for derived morphological forms. Given this model, we are able to predict the types of overgeneralizations that learners make and the way in which these overgeneralizations are dropped; i.e. how learners retreat from overly general rules, without access to negative data. Systems which integrate linguistic theory, learnability models and empirical data about children's grammars allow us to develop theories of language in which issues raised by logical models and experimental findings can challenge and inform one another. This study is an effort toward this kind of integration.

This work is intended as an attempt to bring the phenomenon of resumptive pronouns under the scrutiny of analytical techniques current in generative grammar, particularly the syntactic theories of Government and Binding and Generalised Phrase Structure Grammar. It is claimed that the notion 'resumptive pronoun' finds a definition within generative grammar that is both interesting typologically and theoretically. It is argued that not all apparent instances of resumptive pronouns are truly so, and that languages may differ typologically in whether their grammars countenance 'true' resumptive pronouns or not. Data from Swedish, Hebrew, Irish and Welsh is considered in some detail and it is claimed that there is no universal uniformity in the grammatical devices a language may employ in its system of resumptive pronouns. The consequences of the proposed account of the data from the four mentioned languages for Government-Binding theory and Generalised Phrase Structure Grammar are discussed and integrated into the particular (sub-) theories that relate to them. Semantically, it is claimed that resumptive pronouns show the characteristics expected of pronouns rather than gaps (empty categories); data from Hebrew are presented to show that there are systematic differences in interpretation for constructions depending on whether the construction contains an empty category or a resumptive pronoun. These semantic issues are presented within the framework of Discourse Representation Theory as developed by Hans Kamp. Finally, data from English are presented to elaborate on the semantic nature of resumptive pronouns; it is claimed that English lacks resumptive pronouns and has instead what are dubbed 'intrusive' pronouns. The properties of the interpretation of intrusive pronouns are shown to follow from general and independent principles, providing support for the analysis in terms of Discourse Representation Structures.

This dissertation is a study of the notion of subject in generative grammar. I will explore four distinct areas in syntax where the definition of subject plays a role in determining the opacity domain for NPs and variables. In particular, I wil examine the thematic structure of NPs, reflexives in German infinitivals, Turkish relative clauses, and extractions from adjunct phrases. In this work, I will attempt to characterize the role that subject plays in each of these areas. The major thesis is that there are two predication systems, A-Predication and A'-Predication (where A is argument and A' is non-argument). By adopting a generalized definition of subject in terms of external argument, we hope to provide a unified theory of A- and A'-binding, much as Aoun's theory does for Government-Binding. The resulting theory is a hybrid of Predication Theory and Generalized Binding.

Suppose John introduced Bill and Tom to Sue and performed no other introductions. Then (i) "John only introduced Bill to SUE" is true, while (ii) "John only introduced BILL to Sue" is false, where capitalization symbolizes a focus marked by a phonetic prominence. Two analyses of this phenomenon of association with focus are considered. The scope theory posits a logical form in which the focused phrase and a lambda abstract with a bound variable in the position of the focused phrases are arguments of "only". According to the domain selection theory I propose, (i) and (ii) have a function-argument structure mirroring the syntax. The translation of "only" has two arguments, the VP translation and the translation of the subject NP; (i) expresses a quantification over properties. Focus contributes to the meaning of (i) by delimiting the domain of quantification to properties of the form 'introduce Bill to y', where y is an individual. This yields an assertion: if John has a property of the form 'introduce Bill to y', then it is the property 'introduce Bill to Sue'. This is similar in truth conditions to the assertion produced by the scope theory, namely 'if John introduced Bill to y, then y is Sue'. This idea is executed by including a recursive definition of the sets which will serve as domains of quantification in a Montague grammar. It is argued that the domain selection theory is superior in several ways. In particular, no bound variable in the position of the focused phrase is postulated; the relation between "only" (or "even") and a focused phrase violates structural conditions on bound variables. Chomsky's crossover argument for assigning scope to focused phrases is answered. The proposal is extended to cases where "only" and "even" modify NP and various other categories by means of a crosscategorial semantics analogous to the crosscategorial semantics for conjunction proposed by Gazdar and others. Other constructions discussed are association of focus with adverbs of quantification (MARY always takes John to the movies, Mary always takes JOHN to the movies), clefts (it is JOHN's father who came, it is John's FATHER who came), and conditionals.

This dissertation examines the distribution of high vowels and glides using Prince and Smolensky's Optimality Theory, which proposes that phonological constraints are violable and hierarchically ranked. The distribution of high vowels and glides is shown to be a consequence of simultaneously comparing moraic and nonmoraic syllabifications of high vowels for satisfaction of phonological constraints. In brief, a high vowel surfaces when the moraic parse best satisfies the constraints and a glide surfaces when the nonmoraic parse best satisfies the constraints. This dissertation investigates three main phenomena associated with the distribution of high vowels and glides. First, it treats the syllabification of vowel sequences in a number of languages with only surface monophthongal vowels. In Etsako, Luganda, Kimatuumbi, and Ilokano, high vocoids are syllabified as vowels when followed by a consonant, but they are syllabified as their nonmoraic counterparts when followed by another vowel. Furthermore, the syllabification of nonhigh vowels varies across these languages. The syllabification of vowel sequences is shown to follow from the interaction of syllable structure constraints that ensure the surface vowel is a monophthong. The interlinguistic variation in syllabification is shown to follow from different rankings of the same set of syllable structure constraints. Second, stress can influence the distribution of high vowels. In Lenakel and Spanish, the generalization is that a high vocoid adjacent to a nonhigh vowel is a vowel when stressed otherwise it is a glide. This generalization implies that stress placement must be known prior to syllabification, which is problematic in procedural approaches to constituent construction, where syllabification must precede metrification. In the Optimality-Theoretic approach, the distribution of high vowels is determined by simultaneously best satisfying the metrical and the syllable structure constraints. Third, the distribution of high vowels and glides cannot always be attributed to an alternation between underlying vowels and glides. In a language like Berber, glides must be present underlyingly, and these underlying glides can alternate with high vowels. This is often called glide vocalization. The alternation between underlying glides and high vowels in Berber is also shown to be the result of constraint interaction. In this case, moraic and nonmoraic syllabifications of the underlying glide are compared for constraint satisfaction.

The central claim of this thesis is that direct objects in English move overtly to a Case position external to VP. This proposal differs from the standard GB (e.g. Chomsky 1981) approach in which the verb assigns Case to the direct object within VP. This proposal also differs from the current Minimalist (Chomsky 1993) approach in which the direct object is in VP in the overt syntax only moving to a VP-external Case position at LF. The first four chapters outline this basic hypothesis, concentrating on standard direct objects in English. The evidence is of various sorts: (1) at LF, objects asymmetrically c-command VP and other VP-internal material; (2) overtly the main verb and the direct object are external to a constituent containing the remainder of the VP material, analyzed as VP itself; (3) adverb placement distinguishes between functional and lexical projections providing further support for the present proposal over a VP shell account (Larson 1988). The subsequent two chapters examine other types of "object" construction: ECM and the double object construction. The sorts of evidence adduced for the preceding claims show the ECM subject as well as both objects of the double object construction to appear overtly in VP-external AGRo specifiers, further supporting a Case-checking account invoking functional specifiers. The final two chapters explore the differences between LF and PF NP positions. There various LF "lowering" phenomena are explained by exploiting the copy and delete strategy for movement (Chomsky 1993). This account leads to an examination of copying and deleting and a proposal for the mechanics of such an account. There insertion and stylistic inversion are provided a new account by putting together the parts of the analysis proposed here. An interesting conclusion reached in this thesis is that there is no A-movement at LF in English. In fact, if copying is interpreted as proposed here (following Marantz 1994), there can be no A-movement at LF universally. The differences between pre-SPLIT and LF "movement", then, follow from the different strategies exploited at these levels, only the former exhibiting "reconstruction" effects due to using the copy and delete strategy.

In this dissertation I propose a rank-ordering model of EVAL. This model differs from classic OT as follows: In classic OT, EVAL distinguishes the best candidate from the losers, but does not distinguish between different losers. I argue that EVAL imposes a harmonic rank-ordering on the complete candidate set, so that also the losers are ordered relative to each other. I show how this model of EVAL can account for non-categorical phenomena such as variation and phonological processing. Variation. In variation there is more than one pronunciation for a single input. Grammar determines the possible variants and the relative frequency of the variants. I argue that EVAL imposes a harmonic rank-ordering on the entire candidate set, and that language users can access more than the best candidate from this rank-ordering. However, the accessibility of a candidate depends on its position in the rank-ordering. The higher a candidate appears, the more often it will be selected as output. The best candidate is then the most frequent variant, the second best candidate the second most frequent variant, etc. I apply this model to vowel deletion in Latvian and Portuguese, and to [t, d]-deletion in English. Phonological processing. Language users rely on grammar in word-likeness judgments and lexical decision tasks. The more well-formed a non-word, the more word-like language users will judge it to be. A more well-formed a non-word is considered more seriously as a possible word, and language users will be slower to reject it in a lexical decision task. The rank-ordering model of EVAL accounts for this as follows: EVAL compares non-words and imposes a rank-ordering on them. The higher a non-word occurs in this rank-ordering, the more well-formed it is. Therefore, the higher a non-word occurs, the more word-like it will be judged to be, and the more slowly it will be rejected in lexical decision tasks. I illustrate this by discussing two sets of experiments on how grammar influences phonological processing. The first set investigates the influence of the OCP on processing in Hebrew, and the second the influence of a constraint on [sCvC]-words in English.

For all the progress that has been made in understanding the syntax-semantics interface, many long-familiar connections between the meaning and position of modifiers remain largely mysterious. Why should color adjectives, for example, occupy different positions than size adjectives and evaluative adjectives? Why should subject-oriented adverbs occupy different positions than manner adverbs? More generally, why should the lexical semantics of modifiers often seemingly determine their position in this intricate way? The analytical thread unifying the independent case studies in this thesis is an attempt to address this longstanding question by attributing part of the apparent lexical semantics of certain modifiers instead to their position itself. In Chapter 1, I discern in previous research a common analytical impulse to posit grammatical devices that mediate between a particular modifier and the expression it modifies. I then propose a generalized understanding of how such mediation might work, building broadly on McConnell-Ginet (1982) and Cinque (1999), in which certain modifiers are treated as arguments of semantically-interpreted features in the functional structure of the modified expression. This permits a kind of decomposition, in which a modifier's lexical semantics is, in the spirit of Ernst (2002), reduced to a single simple core meaning that remains constant irrespective of syntactic position, and the semantic peculiarities that vary from one position to another arise instead directly form the denotations of features. After sketching how this might provide some analytical traction on some otherwise vexing puzzles, I move on in other chapters to examine some particular not-well-understood modifiers. Chapter 2 focuses on a class of evaluative adverbs such as remarkably that occur inside the adjectival projection, arguing that they are interpreted as arguments of degree morphology that imputes to them a roughly exclamative-like domain-widening interpretation. Chapter 3 develops an account of the distinguishing properties of weak DP adverbials—low scope, fixed position, and a durativity presupposition—in which they are linked to Aktionsart information in the VP and mirror the syntax of other types of measure phrase. Chapter 4 examines modifiers such as almostand virtually cross-categorially, assigning them an intensional semantics. Chapter 4 analyzes whole and entire as fundamentally non-quantificational expressions that reduce exception tolerance.

In some languages, a subset of the lexicon is exempt from generalizations that hold over the rest of the lexicon. Usually, the words that make up this exceptional class are loanwords from a foreign language. This dissertation argues that such a partitioning of the lexicon more or less according to the historical origins of words is psychologically real to native speakers and should be considered part of Universal Grammar. I conducted a series of experiments using speakers of Japanese, English, and Latvian to show that speakers use knowledge of the stratification of their languages to resolve ambiguities in the sound signal. I found some evidence to suggest that speakers interpret certain marked speech sounds as a sign of foreignness and also that speakers are aware of the foreign origin of some unmarked words. I argue that information concerning the source of unmarked foreign words can only come from non-grammatical sources. However, once supplied with this information, the grammar can account for perceptual bias, as well as classify marked words into their respective lexical strata. These results can be obtained by combining several already existing proposals that have been made recently within the Optimality Theory framework.