Abstract

Belief attributions pose difficult problems for the semantics of natural languages. One problem is to give an account of the "objects" of belief that respects the logical opacity of belief . Another problem is to respect the partial deductive closure of belief: the fact that belief attribution presupposes that believers draw some, but not all, logical consequences from their beliefs. On standard theories, however, partial deductive closure seems to imply total deductive closure. ;I critically examine the major semantic theories of belief attributions--among others those due to Russell, Cresswell, Fodor, and Stalnaker--and argue that their defects are rooted in the framework in which they are constructed. I propose an alternative framework, one that is built on the following two ideas: Objects of belief are not to be identified with the meanings of sentences. I propose that they be identified with "Informational Contents". Informational Contents, as I conceive of them, are abstract, language independent entities whose individuation is sensitive to the logical and semantic complexity of the that-clauses, and to pragmatic and psychological parameters. The partial deductive closure of our beliefs, I suggest, be accounted for by treating 'believe' as a vague predicate--a predicate that is tolerant of marginal changes in the conceptual and logical complexity of its objects. I argue that the problem of partial deductive closure is parallel to that of the Sorites paradox, and should be solved in a parallel way. ;The framework suggested by these two ideas will yield, I believe, an adequate semantic theory of belief sentences, one that reflects the underlying presuppositions of ordinary usage