Explanatory Content and Multi-Turn Dialogues in Tutoring

We aim to systematically investigate the effects of explanatory content and multi-turn dialogues on learning during tutoring by implementing alternative tutoring regimens in an intelligent tutoring system for a letter sequence extrapolation task. It is plausible that the power of tutoring resides, in part, in the fact that a tutor can extend right/wrong discourse moves like "OK" and "are you sure" with explanatory content ("this is right/wrong, because ..."). In past work, we have proposed a computational theory of learning from information about wrong answers (Ohlsson, 1996). However, some empirical tutoring studies have found an effect of explanatory content, others have not. The effects of explanatory content might depend on the linguistic devices used to communicate it. For example, backward references in multi-turn dialogues ("remember what we said before about ....") are part of normal discourse, and evidence suggest that they add pedagogical power. However, resolving such references requires working memory capacity, so they might interfere with learning. A deeper understanding of the interaction of explanatory content and linguistic form is important for determining the optimal design of natural language interfaces for intelligent tutoring systems (DiEugenio, 2001). We investigate this issue by systematically varying the content and the linguistic form of tutoring explanations, and assessing the effects in controlled experiments. The relevant dimensions of variation are identified via the analysis of tutoring dialogues generated by novice and expert tutors.