When Does Disengagement Correlate with Learning in Spoken Dialog Computer Tutoring?

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

• Type: Journal
• Year: 2011
• Where: AIED
• Authors: Katherine Forbes-Riley, Diane J. Litman
• Keywords: AIED 2011, Artificial Intelligence, Dialog Computer, Disengagement, Prerequisite Knowledge