Socially Capable Conversational Tutors Can Be Effective in Collaborative Learning Situations

Rohit Kumar, Hua Ai, Jack L. Beuth, Carolyn P. Rosé

Tutorial Dialog has been shown to be effective in supporting both individual as well as group learners. However, unlike the case with individual learners, teams of learners often ignore and abuse the automated tutors. Both theory and empirical work in the area of small group communication argue that group participants display both task as well as socio-emotional behaviors during interactions. However, in connection with automated conversational agents, the effects of socio-emotional behaviors are much less well understood, especially in the case of multi-party interactions. In this paper, we will describe an evaluation of a socially capable conversational tutor that supports teams of three (or more) learners in a design task. This tutor is evaluated in comparison with a socially neutral baseline agent and human capability “gold standard” tutors demonstrating that the socially capable tutor achieves significantly higher learning gains than the neutral, purely task focused tutor and learning gains not significantly different from the human capability “gold standard” tutors.

The final publication is available at Springer via