Abstract
In this paper, we show our corpus for emergent dialogues. Emergent dialogues have two features, improvisation and creativity. Improvisation means we have to achieve our aim while we are looking for an action required on the spot each other during a dialogue. Although we try to have a dialogue based on a prior plan about the content of the dialogue, we would fail the dialogue. Creativity means we sometimes conceive a new idea after we hear our partner's utterance. To convey the idea to our partner have influence on our partner's thought. These two features are important in human spontaneous dialogues. Our aim is to construct a dialogue management model for emergent dialogue, and to implement it on a computer. We conducted an experiment using tasks in order to collect emergent dialogues. The solutions of the tasks and their correct answers are not clear. We used two experimental settings, “visual and audio” and “audio only”. We report features in relation to how turn-taking were operated and how agreement expressions were used to come to mutual agreement in the dialogue.
