Designing Embodied Conversational Agents
The design of
embodied conversational agents puts many demands on system architecture. Therefore, we describe a conversational framework expressed as a list of conversational properties and abilities and then demonstrate how it can lead to a set of architectural design constraints. We describe an architecture that meets the constraints, and an implementation of the architecture that therefore exhibits many of the properties and abilities required for real-time natural conversation.
We are currently developing an embodied
conversational agent architecture that integrates the real-time multimodal aspects of Ymir with the deep semantic generation and multimodal synthesis capability of Animated Conversation. We believe the resulting system provides a reactive character with enough of the nuances of human face-to-face conversation to make it both intuitive and robust. We also believe that such a system provides a strong platform on which to continue development of embodied conversational agents. And we believe that the conversational framework that we have developed as the underpinnings of this system is general enough to inform development of many different kinds of embodied conversational agents.