Participating in Instructional Dialogues: Finding and Exploiting Relevant Prior Explanations

In this paper we present our research on identifying and model-ing the strategies that human tutors use for integrating previous explanations into current explanations. We have used this work to develop a computational model that has been partially implemented in an explanation facility for an existing tutoring system known as SHERLOCK. We are implementing a system that uses case-based reasoning to identify previous situations and explanations that could potentially aaect the explanation being constructed. We have identiied heuristics for constructing explanations that exploit this information in ways similar to what we have observed in instructional dialogues produced by human tutors. When human tutors engage in dialogue, they freely exploit all aspects of the mutually known context, including the previous discourse. Utterances that do not draw on previous discourse seem awkward, unnatural, or even incoherent. Previous discourse must be taken into account in order to relate new information eeectively to recently conveyed material, and to avoid repeating old material that would distract the student from what is new. Thus, strategies for using the dialogue history in generating explanations are of great importance to research in natural language generation for tutorial applications. The goal of our work is to produce a computational model of the eeects of discourse context on explanations in instructional dialogues, and to implement this model in an intelligent tutoring system that maintains a dialogue history and uses it in planning its explanations. Based on a study of human-human instructional dialogues, we have developed a taxonomy that classiies the types of contextual eeects that occur in our data according to the explanatory functions they serve (16). In this paper, we focus on one important category from our taxonomy: situations in which the tutor explicitly refers to a previous explanation in order to point out similarities (diierences) between the material currently being explained and material presented in earlier explanation(s). We are implementing a system that uses case-based reasoning to identify previous situations and explanations that could potentially aaect the explanation being constructed. We have identiied heuristics for constructing explanations that exploit this information in ways similar to what we have observed in instructional dialogues produced by human tutors. By building a computer system that has this capability as an optional facility that can be enabled or disabled, we will be able to systematically evaluate our hypothesis that this is a useful tutoring strategy.