Implementing analogies in an intelligent tutoring system by modeling human tutoring sessions
Carl Robert Carlson, Martha W. Evens, Evelyn Lulis · 2005
The goals of this research are to understand how human tutors use analogies and to implement them in an intelligent tutoring system called CIRCSIM-Tutor. In pursuit of this goal, human tutoring sessions were studied and analyzed, in regard to the human tutors' use of analogy, and their behavior modeled in our system. The corpus consisted of tutoring sessions regarding the baroreceptor reflex conducted by experts Michael and Rovick. The sessions were marked up, by hand, in an SGML-compliant annotation language, following the approach of Freedman and Kim. New markup were developed to accomplish this analysis. It was observed that, although analogies were not frequently used by the tutors, when they were used, they were very successful. Tutors used analogies to enhance student understanding, to reinforce material learned, and to correct misunderstandings. Analogies were categorized as abstract or concrete, and by goal, base, and target following Gentner's Structure-Mapping approach. They were also categorized as reflective or not, defining reflective analogies as those designed to get the student to think about experiences in the tutoring session and make new generalizations. A series of discussions with Joel Michael led decisions regarding which analogies would be implemented in CIRCSIM-Tutor. While reviewing the analogies in the corpus, common student misunderstandings that were successfully addressed by evoking analogies were identified. Schemas were developed for these analogies. The development of APE operators for implementation in CIRCSIM-Tutor and CAPE is in progress. The Turn Planner for Version 3 of CIRCSIM-Tutor was integrated in the system and altered to deploy analogies into the turn structure. The system's Discourse Planner was modified to call the analogy operators.