Depends on Who Said It: Talker-sensitive Inference of Pragmatic Meaning from the Acoustic Signal

Spoken language communication in English proceeds at the rate of ~2.5 words per second. The mechanism through which human listeners derive contextually enriched inferences from the rapidly unfolding input has been a topic of importance in psycholinguistics (Degen & Tanenhaus, 2016; Grodner et al., 2008; Huang & Snedeker, 2019; Sedivy et al., 1999; inter alios). Often implicit in this inquiry is the assumption that listeners are able to seamlessly map the continuous acoustic signal to linguistically meaningful, discrete representations (e.g., phonemes, words, intonation contours). The mapping is, however, flexible and often variable across talkers and contexts, creating ubiquitous ambiguity in the inference process. In this talk, I will argue that listeners navigate the variability, at least partially, by drawing on covariance relationships between the linguistic input and socio-indexical features of talkers. By examining comprehension of scalar adjectives (Dix et al., under review; Ryskin, Kurumada, & Brown-Schmidt, under review) and intonational prosody (Buxó-Lugo & Kurumada, under review), I demonstrate that 1) listeners are sensitive to the underlying structure of the across-talker variability in multiple domains of the linguistic signal; and 2) smooth and accurate language comprehension draws on adaptive inferences listeners make regarding the signal-meaning mapping that is most likely given characteristics of a current talker.