Affect expression in ECAs: Application to politeness displays

In this paper we present our Embodied Conversational Agent (ECA) capable of displaying a vast set of facial expressions to communicate its emotional states as well as its social relations. Our agent is able to superpose and mask its emotional states as well as fake or inhibit them. We defined complex facial expressions as expressions arising from these displays. In the following, we describe a model based on fuzzy methods that enables to generate complex facial expressions of emotions. It uses fuzzy similarity to compute the degree of resemblance between facial expressions of the ECA. We also present an algorithm that adapts the facial behaviour of the agent depending on its social relationship with the interactants. This last algorithm is based on the theory of politeness by Brown and Levinson (1987). It outputs complex facial expressions that are socially adequate.