We Click, We Align, We Learn

Behavioral convergence has been identified as one (largely subconscious) contributor to successful conversations, while rapport is one of the central constructs that explains development of personal relationships between these speakers over time. Social factors such as these have been shown to play a potent role in learning. Therefore, in this work, we investigate the relationship in dyadic peer tutoring conversations of convergence, building up of interpersonal rapport over time, and student learning, while positing a novel mechanism that links these constructs. We develop an approach for hierarchical computational modeling of convergence by accounting for time-based dependencies that arise in longitudinal interaction streams, and can thus a)quantify the effect of one partner's behavior on the other and differentiate between driver and recipient (Influence), b)extrapolate the outcome of directional influence to determine adaptation in partners' behaviors (Convergence). Our results illustrate that influence, convergence and rapport in the peer tutoring dialog are correlated with learning gains and provide concrete evidence for rapport being a causal mechanism that leads to convergence of speech rate in the interaction. We discuss the implications of our work for the development of peer tutoring agents that can improve learning gains through convergence to and from the human learner's behavior.