Implementing analogies in an electronic tutoring system

Abstract

We have built an ITS system for cardiovascular physiology that carries on a natural language dialogue with students and simulates our expert tutors' behavior. The tutoring strategies and language are derived from a corpus consisting of eighty-one hour long expert human tutoring sessions with firstyear medical students at Rush Medical College. In order to add analogies to the available tutoring repertoire, we analyzed the use of analogies in the human tutoring sessions. Two different types of analogies were discovered:one involves reflection on students' earlier work and the other refers to familiar things outside the physiological domain, like balloons and Ohm's Law. The two types involve different implementation approaches and different language. We are now implementing analogies of the first type in our ITS using the same schemas and rule-based Natural Language Generation techniques as in the rest of the dialogue that CIRCSIM-Tutor generates. We are using the Centner's model of analogy and Forbus' Structure Mapping Engine to implement the second type. © Springer-Verlag 2004.