Abstract

We will present on-going work on endowing virtual agent with soft conversational skills. We will concentrate on multimodal nonverbal behaviours, for the locator and the interlocutor within a dyadic interaction. We will present a real-time platform of agents that exhibit backchannel signals in relation to their interlocutors (be virtual or real (human user)). We will illustrate our topic with examples from the EU-project SEMAINE where we have developed 4 agents called Sensitive Artificial Listeners able to engage the user in a conversation. We will present another study we are conducting dealing with smile. Smile may take different meanings depending on its morphological properties and interaction context. We consider 3 types of smile: happy, polite and embarrassed. To find out about smile morphological characteristics, we asked participants to create interactively these smiles onto a virtual agent. The results of this study allowed us also to enlarge the repertoire of behaviors of the agent.