Dynamic speech adaptation to unreliable cues during intonational processing

Abstract

Human behavior is often remarkably flexible, showing the ability to quickly adapt to the statistical peculiarities of a particular local context. When it comes to language, previous work has shown that listeners' anticipatory interpretations of intonational cues are adapted dynamically when cues are observed to be stochastically unreliable. This paper reports novel empirical data from manual response dynamics (mousetracking) on how listeners adapt their predictive interpretation when some intonational cues are occasionally unreliable while others are consistently reliable. A model of rational belief dynamics predicts that listeners adapt differently to different unreliable intonational cues, as a function of their initial evidential strength. These predictions are borne out by our data.

Mapping prosody onto meaning – the case of information structure in American English

“…Listeners' ability to make use of bottom-up acoustic cues may be complemented by probabilistic knowledge about speaker production likelihoods, i.e. how likely the speaker is to use a particular prosodic form in order to express a particular discourse function (Buxo-Lugo, 2017;Buxo-Lugo & Watson, 2016;Kurumada et al 2014, b, Roettger & Franke, 2018a.…”