Bridging the Gap Between Empirical Data on Open-Ended Tutorial Interactions and Computational Models

In this paper we present an approach to using empirical data on human teacherlearner interactions to guide the development of a pedagogical agent for supporting musical composition learning. Our approach to bridging the gap between tutorial interaction analysis and computational models, intended for use in learning support systems, is a new one. We support our claim by pointing out that most of the previous work in the area of using human tutors as models has been conducted in domains that are more procedural than the open-ended subject area investigated here. However, the approach described in this paper seems applicable to most, if not all, domains, whether open-ended or not. In the paper we describe how an empirical study of teacher-learner interactions was linked, specifically modulated, to the construction of a pedagogical agent called MetaMuse. Empirically derived state transition networks were used to provide a semi-open, goal-oriented interaction plan. One distinctive feature of MetaMuse is its use of student input, regarding whether their expectations were met, to stimulate the pedagogical agent into action. The paper concludes with a discussion of the utility of our approach.