Dialogue Behavior Control Model for Expressing a Character of Humanoid Robots

Abstract

This paper addresses character expression for humanoid robots that play a social role via spoken dialogue so that the character matches to the given social role such as a lab guide or a counselor. While conventional methods of character expression mostly focused on changing the style of utterance texts, this study focuses on dialogue behavior features that may affect the impression of spoken dialogue. Specifically, we use five dialogue behavior features: utterance amount, backchannel frequency, backchannel variety, filler frequency, and switching pause length (the time until the system responds). We adopt three character traits of extroversion, emotional instability, and politeness for character expression. We then investigate the relationship between the dialogue behavior features and the character traits by conducting subjective evaluations. A statistical analysis of the subjective evaluations shows that the dialogue behavior features except for the backchannel variety are related to either of the character traits. By using the subjective evaluation scores on the relevant traits, we can train models to control the dialogue behavior features of a robot according to the desired character. Another experimental evaluation demonstrates the feasibility of character expression with regard to the traits of extroversion and politeness.