Abstract

In this dissertation I argue that truth-conditional semantics for vague predicates, combined with a Bayesian account of statistical inference incorporating knowledge of truth-conditions of utterances, generates false predictions regarding negations and metalinguistic inference. I thus propose a fundamentally probabilistic semantics for vagueness on which the meaning of a vague predicate is a likelihood function on the states it encodes, with these likelihoods being generated via reinforcement learning in a signaling game.