Abstract
In this paper, we describe a dialogue-based system that utilizes a teacher-configurable conversational agent to trigger students' discourse in collaborative activities. This prototype web-based system, named MentorChat, aims to induce learning-generating cognitive activity by prompting online learners' discussions. We evaluated the effect of the conversational agent on student learning in a collaborative context by conducting an experimental lab study. Two conditions were contrasted: (1) students who discussed online and provided an answer to a key domain issue guided only by a collaboration script (control condition), and (2) students who, along with the collaboration script, were prompted by the agent during their discussion, to elaborate on key domain concepts (treatment). Our findings suggest a positive learning effect of the treatment condition. They also indicate that the agent unsolicited prompts can improve students' ability to recall important domain concepts in a post-test control instrument.