Expression of affects in embodied conversational agents

Providing the virtual agents with expressive faculties is a contemporary challenge. The generated emotional behaviour for Embodied Conversational Agents (ECAs) particularly has to be credible and varied. Cowell and Stanney (2005) showed that well chosen behaviours and emotional expressions improve the credibility of an agent. Behaviour expressivity can be conveyed through: the choice of the nonverbal signals and their realisation. Some researchers argue that creating an “internal state ” for an ECA and giving the ECA the possibility to express emotions contributes to a richer human-machine interaction (Walker, Sproull and Subramani 1994). The internal states can be expressed on the behaviour level through the choice of the nonverbal signals and the modulation of their realisation. In this Chapter several algorithms that enrich the behaviour repertoires of ECAs are presented. Although a great number of models use interpolations to create new expressions based on formerly defined basic emotions (e.g. Tsapatsoulis et al. 2002; Albrecht et al. 2005; Paradiso 2002), a particular attention is paid to the models of emotional expressions going beyond these so-called universal expressions. Some computational models of perceptively verified expressions have been created, mostly using fuzzy methods (Arya and DiPaola 2007