Whamit!

The standard Hintikkan semantics views “believe” as a universal quantifier over possible worlds, stating that the prejacent is true across all the attitude holder’s doxastic alternatives (Hintikka 1969). However, this semantics (i) fails to capture the fact that “believe” is a gradable predicate (e.g. “partially believe” v. “fully believe”) and (ii) makes no predictions about the degree of certainty of the belief agent. To remedy these problems, I propose a probabilistic semantics along the general lines of Kennedy & McNally’s (2005) analysis of gradable adjectives. I argue that “believe” is a strong modal, i.e. it is a maximum-degree predicate. While belief attributions can sometimes be interpreted as hedges (e.g. “I believe it’s raining, but I’m not sure it is”), I argue (contra Hawthorne et al. 2016) that such weak uses are not the default as they canonically arise with first-person present-tense unembedded forms and under the right pragmatic conditions, i.e. when the belief component is not relevant to the question under discussion. Following up on a suggestion made in Chemla (2008), I propose that the weak sense of “believe” arises as an antipresupposition, i.e. as a scalar inference derived through competition with a presuppositionally stronger “know”-competitor. A weak interpretation amounts to a situation in which the speaker expresses full subjective confidence in the prejacent but reneges on publicly committing to it.