When does disengagement correlate with learning in spoken dialog computer tutoring

摘要: We investigate whether an overall student disengagement label and six different labels of type are predictive learning in a spoken dialog computer tutoring corpus. Our results show first that although students' percentage disengaged turns negatively correlates with the amount they learn, individual types correlate differently learning: some learning, while others don't at all. Second, we these relationships change somewhat depending on prerequisite knowledge level. Third, using multiple to predict improves power. Overall, our suggest adapting should improve maximizing requires system interventions type.