From observation to interaction

Conversation is a natural and powerful means of communication for people to collaboratively create and share information. People are very fluent both in aligning their behavior with each other and in coordinating multiple modalities to create meaning. A big challenge is to build an intelligent robot that can participate in conversation so as to mediate the knowledge process in a community.

In this talk, I present an approach to reproducing a competence of participating in conversational interactions in a data intensive fashion. First, I describe a method of measurement and analysis for gaining an in-depth quantitative understanding of conversation. Second, I present a suite of learning algorithms for enabling our robot to build and revise a competence of communication as a result of observation and experience. Third, I discuss a conceptual framework of social intelligence design in which conversation is characterized in a larger loop of information and knowledge circulation. Finally, I discuss the current limitations and future work.