Constructing Knowledge from Dialog in an Intelligent Tutoring System: Interactive Learning, Vicarious Learning, and Pedagogical Agents

摘要

College students either interacted directly with an intelligent tutoring system, called AutoTutor, by contributing to mixed initiative dialog, or they simply observed, as vicarious learners, previously recorded interactive sessions. The mean pretest to posttest effect size (Cohen's d) across two studies was 1.86 in the interactive conditions and 1.12 in standard vicarious conditions. In Experiment 1, redundant onscreen printed text produced an effect size of 0.43, but the difference was not significant. In addition, the image of a talking head presenting AutoTutor's contributions to the dialog while displaying facial expressions, gestures, and gaze did not produce learning gains beyond those produced by the voice alone. In Experiment 2, the effect size was 0.71 when interactive tutoring was contrasted with the standard vicarious condition, but only 0.38 when compared to a collaborative vicarious condition.