Greta, an Interactive Expressive Embodied Conversational Agent

Greta is an interactive Embodied Conversational Agent platform. It is endowed with socio-emotional and communicative behaviors. Through its behaviors, the agent can sustain a conversation as well as show various attitudes and levels of engagement. Through the years, we have integrated all our research in the Greta platform. By applying different methodologies, based on corpus analysis, user-centered, or motion capture, we have enriched the agent's palette of multimodal behaviors. We have conducted various studies to simulate communicative behaviors, emotional behaviors, social attitudes and behavior expressivity. In particular we have proposed models to go beyond the prototypical expressions of emotions. Through its behaviors patterns, the agent can display complex emotions such as masking one expression of emotions by another ones, its relationship towards its interlocutors, specific social signals such as smile and laughter. In an interaction, the agent can be a speaker or a listener. It can exhibit backchannels, mimic on the fly its interlocutor's behaviors. To develop our models, we rely on theoretical models from social psychology literature and on data analysis. After describing our platform, we will first review our rationale; then we will introduce our model of socioemotional behaviors. Finally we will present experiments where we measure the impact of the agent's copying behaviors on the user's level of engagement.