Toward constructing a dialogue management model for Kyowa

Utterances in a Japanese dialogue are usually overlapped by Aiduchi (back channel), making up for another´s words, and so on. Such a dialogue is called Kyowa (collaborative dialogue) in Japanese. In Kyowa, two persons utterances constitute a single sentence. We have to construct a sentence in Kyowa collaboratively. We make our dialogue Kyowa for the purpose of enhancing a social human relation rather than insisting on our own opinion. Our aim is to construct a Kyowa-type dialogue management model and to implement it on a computer. We conducted an experiment in which the subjects´ dialogue was expected to be of Kyowa-type, and we collected the dialogues. We used two experimental settings, “visual and audio” and “audio only”, and here we report features found when analyzing these dialogues