Exploring the Difference of the Impression on Human and Agent Listeners in Active Listening Dialog

Abstract

Active listening is a communication technique that the listener listens to the speaker carefully and attentively by confirming or asking for more details aboutwhat they heard. In order to improve the effect, always-available and trustable conversational partners in enough number are demanded. The ultimate goal of this study is the development of a virtual agent who can engage active listening and maintain a long-term relationship with elderly users. We assume that the task of the active listener (a human volunteer or the agent) is to maintain the speaker’s (elderly user) mood in good state. In order to do this, like a human listener, the listener agent has to observe the listener’s attitude, has to estimate the listener’s mood from the observation, and has to predict the change of listener’s mood caused by his / her own behaviors both verbally and non-verbally. On the other hand, the active listener is evaluated by the speaker from his / her impression of the listener’s attitude. The hypothesis is, if the impression is good, then the speaker’s mood is good. However, virtual agents which are made by computer graphics animations are more limited in expressiveness than human listeners, both in the aspects of quality and communication channels. Therefore, there is a research question that the graphical agent with “reduced expressiveness” can really engage the active listening task at human listeners’ level, even if they do the same behaviors, smiles, gestures, or utterances at the same timings. This paper presents our first step of this study, a human-human teleconferencing experiment to foresee whether it is possible to implement an active listener agent.