Conversational Tutors with Rich Interactive Behaviors that support Collaborative Learning

In addition to presenting relevant instructional content to learners in collaborative learning settings, intelligent conversational tutors must be good communicators to keep the students attentive and engaged. In our recent work, we have shown that tutors capable of performing both task-related (tutoring) as well as social behavior are more effective tutors than tutors that perform only task-related behavior. It is also important to consider that collaborative learning tasks in ill-defined domains (like engineering design) require students to negotiate among competing goals. Thus, automated tutors must be capable of supporting students with different roles and potentially competing goals. In this paper, we will describe the most recent implementation of the CycleTalk tutor using the Basilica architecture that demonstrates these social behavior and role supporting capabilities. Also, we briefly discuss results of an experiment using this tutor that optimizes the amount of social behavior in such learning settings.