Intensification of Group Knowledge Exchange with Academically Productive Talk Agents

In recent years, intelligent conversational agents have been used with some level of effectiveness as dynamic support for collaborative learning in online chat. The classroom discourse community offers insights from analysis of effective classroom discussion facilitation practices that might productively inspire the design of such facilitator agents. In this paper, we evaluate one such conversational agent-as-facilitator design, drawn from the literature on what has been termed Academically Productive Talk. Specifically we evaluate the effect of a facilitation strategy referred to as Agree/Disagree, where students are prompted to evaluate the assertions of a partner student. In a simple two condition study, we evaluate the effect of this facilitation strategy in comparison with an otherwise identical condition where this facilitation strategy is absent. The results demonstrate a marginal positive effect on learning (effect size .55 standard deviations) and a significant intensification effect on the collaborative discourse.