Abstract
This paper proposes a novel turn-taking model in which participants in conversations estimate next possible speaker based on the 'preceding utterance attitudes' (want to speak/let someone speak) in advance of actual turn-takings and shows that the estimation model contributes to understanding of turn-taking phenomena. In the existing turn-taking rule of multiparty conversation proposed by Sacks explains that turn-takings are performed by using 'current speaker selects next' technique. In this research, first, preceding utterance attitudes are evaluated by three evaluators by observing conversation scenes and evaluated attitudes are compared with actual turn-taking results. From the analysis, we suggest that the next speaker is selected by expressing and estimating attitudes of utterance among participants rather than others' utterances.
