Abstract
This paper proposes a human-computer interaction system that the computer agent has an emotion calculation process in order that the user feels familiar to the agent. The interaction system is developed based on a speech interface system “Takemaru-kun” constructed by Shikano et al. 4465 question-example texts are prepared for Takemaru-kun and all the texts are determined suitable response texts and animation gestures. When speech recognition result is inputted, the most similar question-example text is chosen and its corresponding response and gesture are presented to the user. However, the Takemaru-kun system always replies the same reaction for the same input. It makes us feel monotonous and mechanical. The variation of reaction is one of the factors to be able to decrease such images. In order to reply different response for same input and different mood, a set of response text is prepared for each mood. Next mood is calculated from present mood and emotions aroused from input utterance. The emotion is calculated from events which are recognized when human hears input utterance and reply response at the present mood. Emotion Generating Calculation method is used to calculate emotions. Then, the mood transits on Mental State Transition Network based on aroused emotions and the transition cost. 41 subjects evaluated the impression of proposed system using 18 adjectives. As a result, average of “preference” factor of proposed system was higher than that of non-emotion system. And the averages of “mischievous,” “light-minded,” and “entertain” were also high.
