Rationalizing evaluativity

Abstract

Many degree constructions are evaluative in the sense that they require a measure of some sort to exceed a contextually-determined norm. We assume that this inference is always an implicature (Rett, 2015b), and we develop a game-theoretic pragmatic treatment to explain how and when it arises. Our analysis is couched in a Rational Speech Act (RSA) model of communication, building on the theory of vagueness-resolution proposed in Lassiter and Goodman 2014. To extend the Lassiter and Goodman model to a wider range of degree constructions, we do two things: ?rst, we incorporate insights from Barker (2002) about the role of comparison-class uncertainty in the interpretation of gradable predicates, and second, we adapt an independent RSA model of Manner implicature (Bergen et al., 2016) to capture the effects of linguistic markedness in degree constructions (Rett, 2015b). Combining these pieces provides the ?rst gradient model of evaluativity inferences, and we conclude by discussing some of its novel predictions.