The Many Faces of Semantic Compositionality
Dissertation, University of California, Riverside (
2000)
Copy
BIBTEX
Abstract
It has long been claimed that if the meaning of a sentence did not depend on the meaning of its parts, and its form---i.e., if language did not conform to the Principle of Compositionality---we could not explain some of our most remarkable linguistic abilities, indeed we would not be able to learn a language. ;The Principle of Compositionality expresses a powerful but vague idea. We can distinguish two different intuitions underlying the idea. One is that the meaning of a complex expression is itself complex, a whole composed of the meanings of the parts of the sentence. The other, which I call 'functional', is that the meaning of a complex expression is determined in a rule-governed way. This is what we mean, usually, when we say it is a function of the meaning of the parts and the structure. ;After distinguishing these intuitions I focus on the second one, which is best developed in the traditional truth-conditional semantics for formal languages. I argue for two theses. The first is that if we conceive of natural language as functionally compositional, in the way formal languages are, we ought to characterize its compositionality differently from how it is usually done: we ought to say that natural language is compositional because its sentences generate schematic or generic semantic information purely in virtue of their form . I show that this characterization renders ineffective a type of arguments against compositionality that exploit well-known facts about belief reports. ;My second thesis is that the claim that natural language is functionally compositional is vulnerable to a new line of criticism. Functional compositionality commits us to the view that all predicates of the same form as 'is blond' perform the same function, for example that they assert possession of a property, and that all sentences of the same form as 'Tom is blond' are evaluated by the same semantic rule, for example by the rule: ;'Tom is blond' is true iff the referent of the singular term possesses the property expressed by the predicate. ;I argue that this is a constrictive view, that other predicates perform different functions and the same rule does not fit all sentences of the same form as 'Tom is blond'. The constriction can be avoided by changing or refining the formulation of the rule, which amounts to producing different compositional theories of natural language. I describe some of these refinements and the assumptions on which they depend, and show that some of them do not work while all take us away from an intuitive idea of compositionality