Generating distinctive behavior for Embodied Conversational Agents

In this paper we describe an algorithm for generating distinctive behavior for Embodied Conversational Agents. To this aim, we introduce the concepts of agent's general behavior tendency, named Baseline, and local behavior tendency, called in turn Dynamicline. Depending on the communicative intentions of the agent, the Baseline is modulated. The obtained behavior tendency corresponds to the Dynamicline which is then used to determine the nonverbal signals and their expressivity the agent will produce to communicate its intentions. We also propose a system to extract the movement expressivity of a human user standing in front of a camera. The extracted characteristics are then used to characterize the agent's Baseline. We end the paper by presenting an evaluation study of our model. © 2010 OpenInterface Association.