Turn-taking in Mandarin Dialogue: Interactions of Tone and Intonation

Fluent dialogue requires that speakers successfully negotiate and signal turn-taking. While many cues to turn change have been proposed, especially in multi-modal frameworks, here we focus on the use of prosodic cues to these functions. In particular, we consider the use of prosodic cues in a tone language, Mandarin Chinese, where variations in pitch height and slope additionally serve to determine word meaning. Within a corpus of spontaneous Chinese dialogues, we find that turn-unit final syllables are significantly lower in average pitch and intensity than turnunit initial syllables in both smooth turn changes and segments ended by speaker overlap. Interruptions are characterized by significant prosodic differences from smooth turn initiations. Furthermore, we demonstrate that these contrasts correspond to an overall lowering across all tones in final position, which largely preserves the relative heights of the lexical tones. In classification tasks, we contrast the use of text and prosodic features. Finally, we demonstrate that, on balanced training and test sets, we can distinguish turnunit final words from other words at ≈ 93% accuracy and interruptions from smooth turn unit initiations at 62% accuracy.