In this paper, we argue that Embedded Conversational Agents (ECAs) need cognitive credibility to usefully participate in mixed communities. One critical aspect of this credibility has to do with the argumentative capacity of ECAs. We believe that socio-cognitive models of interaction can provide a helpful foundation that agents can use to represent, manipulate, and reason about "what is going on" in a collective in order to better engage in ongoing interactions. Furthermore, because people interact in a large proportion using natural language, it is critical that agents be able to directly process language and construct their model from it. We propose a way to build such a model based on empirical data, by extracting and representing patterns of interaction from large archives of online distributed collectives such as the Free/Open-Source Software project Mozilla.
展开▼
