Expressive Multimodal Conversational Acts for SAIBA Agents

We discuss here the need to define what we call an agent conversational language, a language for Embodied Conversational Agents (ECA) to have conversations with a human. We propose a set of Expressive Multimodal Conversation Acts (EMCA), which is based on the Expressive Speech Acts that we introduced in a previous work, enriched with the multimodal expression of the emotions linked to these acts. We have then implemented these EMCA for SAIBA-compliant ECA, and specifically for Greta. We were able to use Greta in experiments aimed at assessing the benefits of our language in terms of perceived sincerity and believability of an ECA using it to interact with a human user.