Excavating semantics

The generative framework for linguistic study (e.g., Chomsky 1957, 1980, 1986) recognizes two types of linguistic knowledge: knowledge of language which abstracts away from language use, called grammar, and knowledge of language pertaining to language use, called pragmatics. This thesis examines two consequences of drawing such a distinction in linguistic knowledge. The first half of the thesis examines the domain of grammatical knowledge. According to Chomsky and many other generative linguistics, the domain of grammar is the sentence. I present evidence which invites a revision of this position. Considering a specific problem in discourse, the interpretation of pronominal reference in discourses such as those in (1), I argue for grammatical knowledge governing certain aspects of discourse construction. (1) a. A dog walked in. It lay down under the table. b. Every worker stopped at the picket line. They denounced the company\u27s policies. In both of the discourse segments in (1), the pronouns refer to discourse referents introduced as an automatic consequence of the preceding quantified noun phrase (Evans 1977, 1980; Heim 1982; Karttunen 1974, 1976). The singular pronoun in (1a) refers to the dog introduced by virtue of the indefinite, a dog. Similarly, the plural pronoun in (1b) refers to the collection of workers introduced by virtue of the expression every worker. These connections are thought to engage pragmatic rather than grammatical knowledge. Drawing on arguments from modularity and language acquisition, I show that a pragmatic analysis of these facts is both insufficient and unnecessary.^ The second half of the thesis addresses a methodological problem arising from the distinction between grammar and pragmatics. Although generative researchers have made a conscious decision to study principles of grammar and principles of pragmatics separately, the evidence on which linguistic theories are based is not so neatly divided. Linguistic evidence must, of necessity, be obtained through channels of language use. Consequently, the available evidence reflects contributions of both grammatical and pragmatic resources. Inferring grammatical knowledge from this composite evidence is a hazardous enterprise, particularly in the study of semantics. Concentrating on the semantics of quantification, I advance a systematic process for extracting truth-conditional semantics from the available sources of semantic evidence.