Computer Generation of a Tutorial Dialogue

The main focus of our work at Rush has been on the taping of human- to-human tutoring sessions and the analysis of the transcripts of those sessions with regard to both language and content. The language analysis involves study of vocabulary, syntax, semantics, and discourse for use in the understanding of ill-formed input and the generation of responses. The anlaysis of content focuses on tutoring rules, modeling rules, and the underlying knowledge used by the tutor. At IIT we are working on the development of the lexical functional grammar rules and lexicon for the cardiovascular sublanguage, on the development of a cardiovascular knowledge base for the tutoring system, on a Prolog prototype for the tutor we are trying to build, and on the initial plans for a tutoring system in Lisp to support our research in text generation. The context of our work is an attempt to build an intelligent version of a remarkably effective conventional tutoring program, CIRCSIM, designed and implemented in Basic. CIRCSIM is intended to tutor first year medical students in the complex negative feedback processes used by the human body to maintain blood pressure.