Leveraging Conversational Agents and Concept Maps to Scaffold Students´ Productive Talk

Over the past decades, intelligent conversational agents have been repeatedly used as pedagogical tools for fostering students´ motivation and engagement in individual learning settings. However, in recent years, the uses of conversational agents have also been explored in collaborative contexts under the scope of Academically Productive Talk (APT). The theory of APT, which has derived from the work of the classroom discourse community, suggests a number of facilitative talk moves that have proven to effectively promote productive classroom conversations. Following the APT perspective, we propose the design of a teacher-configurable conversational agent system, named MentorChat. This dialogue support system leverages concept mapping as a domain modeling tool that enables the implementation of well-targeted agent interventions, aiming to stimulate constructive peer interactions in collaborative learning settings. In this paper, we also present the results of an evaluation study, where 40 students collaborated in dyads in order to accomplish two different tasks in MentorChat. The preliminary findings are quite encouraging as for the ease of use and user acceptance of the proposed conversational agent system.