Cohesion, Semantics and Learning in Reflective Dialog

A corpus of reflective tutorial dialogs was tagged for cohesive relationships between student and tutor. We describe our tagging scheme, and show that certain cohesive features of tutoring dialog are correlated with learning in our corpus. In particular, our semantic cohesive relationship tags are significant predictors of learning, while our lexical tag is not. We find that “abstractive” dialog moves, in which the student or tutor repeats the other’s previous utterance but at a greater level of generality, are significant positive predictors of learning. We also find that tutor moves which repeat the student’s previous utterance but in a less abstract way predict learning in our corpus. These findings suggest that tracking student dialogue moves can enhance student modeling and guide planning of effective natural-