Semantic Adaptation in Gradable Adjective Interpretation

Ming Xiang, Alex Kramer, Christopher Kennedy

Previous studies on learning and adaptation have largely focused on speech perception and syntactic parsing, but much less is known about whether and how language users adjust their semantic representation after being exposed to other individuals' utterances. The current study focuses on the interpretation of gradable adjectives --- expressions with highly context-dependent interpretations --- and investigates how individuals adjust their thresholds of application after exposure to utterances of the same expressions by other language users. Three experiments provide novel evidence to support robust and rapid semantic adaptation for gradable adjectives, the effect of which is modulated by the types of utterances and the class of adjectives participants were exposed to, as well as the communicative goal of the linguistic task and the identity of the communicative partner. We propose a single unified probabilistic belief update analysis to account for all of the observations. Under our account, threshold adaptation naturally falls out as the result of a listener-speaker coordination process, which is guided by general principles of pragmatic reasoning. The current empirical findings and theoretical proposals also find parallels in the perceptual learning and speech adaptation domain, suggesting a domain general mechanism of learning and adaptation at multiple levels of linguistic representation.