Encoding and decoding of meaning through structured variability in intonational speech prosody

Abstract

Speech prosody, the rhythm and intonation in particular, plays an important role in communication of meaning. Rising vs. falling intonation contours signaling the speaker’s indented communicative meanings (i.e., asking a question vs. making a statement) has been widely recognized as a primary example of such prosody. However, what appears to be a straightforward mapping between acoustic features of prosody and hypothesized meanings in fact presents a challenge to the human perceptual and computational mechanisms. Perceptible features of prosody vary across contexts (e.g., talkers), creating ubiquitous ambiguity in the mapping. Here, we first characterize the structured nature of the variability in intonational speech prosody used to signal a question vs. a statement in English. We then demonstrate that the listener can learn to adapt their expectations about the prosody-meaning mapping according to an inferred underlying structure of the environmental input. We argue that the rich and dynamic representations of the prosodic input allow listeners to infer the mapping that is most likely given characteristics of a current context.

Contour clustering: A field-data-driven approach for documenting and analysing prototypical f0 contours

“…More in general, the literature has shown no consensus on how exactly prosody contributes to the meaning of an utterance (e.g. Couper-Kuhlen 1986, Hirst 2005, Buxó-Lugo & Kurumada 2019. This is partially the result of acknowledged challenges in the field, for example, keeping form and meaning separate in prosodic analyses (Hirst 2005), distinguishing different types of meanings (Cruttenden 1997, Prieto 2015, and accurately modelling prosody beyond intonation (Niebuhr & Ward 2018).…”

“…In this respect it is still largely unclear how prosody and meaning are related (e.g. Buxó-Lugo & Kurumada 2019). A core element in the description of intonation patterns that are representative for a given language is the assumption of an inventory of prototypical tonal patterns (e.g.…”

Listeners’ adaptation to unreliable intonation is speaker-sensitive

“…Buxo-Lugo & Kurumada, 2019, Kurumada et al, 2014b, 2017, Roettger & Franke, 2019. Buxo-Lugo and Kurumada (2019), investigated American English echo questions and statements which can be solely differentiated by the final pitch movement in the utterance. The authors created a continuum between two naturally produced contours and exposed different groups of listeners to different distributions of these contours mapping onto the two discourse functions.…”