A Study of Turn-Yielding Cues in Human-Computer Dialogue

Abstract. Previous research has made significant advances in under-standing how humans manage to engage in smooth, well-coordinated conversation, and have unveiled the existence of several turn-yielding cues — lexico-syntactic, prosodic and acoustic events that may serve as predictors of conversational turn finality. These results have subse-quently aided the refinement of turn-taking proficiency of spoken dia-logue systems. In this study, we find empirical evidence in a corpus of human-computer dialogues that human users produce the same kinds of turn-yielding cues that have been observed in human-human interac-tions. We also show that a linear relation holds between the number of individual cues conjointly displayed and the likelihood of a turn switch.