Model theoretic semantics seeks to understand how syntactic categories and subcategories of words are mapped to certain types of meanings. This relationship between syntax and semantics is particularly relevant for theories of language acquisition (viz., syntactic bootstrapping) and category coercions (e.g., using nouns as verbs or verbs as nouns). This talk discusses not only how facts concerning syntactic bootstrapping and category coercion might influence our semantic theories, but also how our semantic theories might lead to new hypotheses about language acquisition. Both of these issues are explored through an analysis of gradable adjectives.
Gradable adjectives such as `tall’ often presuppose that the elements in their domain are linearly ordered with respect to some gradable property (such as height). For example, competent speakers know that if (1a) is true then (1b) must also be true.
(1) a. John isn’t taller than Bill and Bill isn’t taller than John.
b. John is as tall as Bill and vice versa.
Such conclusions can be reached without knowing anything about the context of utterance, indicating that this entailment relationship follows, at least partly, from the meaning of `tall’ rather than from world knowledge. Interestingly, this presumption of linearity influences how people treat unknown words and even partly determines the interpretation of non-gradable adjectives when they appear with gradable syntax (such as `very,’ as in `Mary is very pregnant’). Generally, new or coerced words that appear in gradable syntax are assumed to linearly order their domains.
In contrast to `tall,’ certain gradable adjectives — such as `successful,’ `good,’ and `smart’ — do not imply that the elements in their domain are linearly ordered. The challenge for an adequate semantic theory is to explain why some adjectives require linear orders and others do not, yet also explain why novel words and usages are presumed to linearly order elements in their domain.
The theory presented in this talk attempts to meet this challenge by hypothesizing a link between context sensitivity and linear orders. It is argued that non-linear orders are derived from gradable adjectives that are underspecified in terms of how they compare and rank individuals on a scale. However, if this hypothesis is correct, then it must be assumed that children and adults presuppose that novel words are not context sensitive (until positive evidence indicates otherwise).