Relationship between Tutorial Goals and Sentence Structure in a Corpus of Tutoring Transcripts

Ninth Midwest Artificial intelligence and Cognitive Science Conference

Volume

Ninth Midwest Artificial intelligence and Cognitive Science Conference

Contents

An intelligent tutoring system, CIRCSIMTutor tutors first-year medical students on blood pressure regulation based on the dialogue patterns of human tutors. To obtain data about the language and conversation patterns of human tutors, we analyzed transcripts of human tutors working over a modem, then annotated them to show tutorial goal structure. In this paper we analyze clusters of sentences serving the same tutorial goal. We attempt to determine the information content required by each group and possible sources of these content elements. We show potential surface structures which could be generated from these elements. We discuss the influence on our work of the theories of Michael Halliday and Deborah Schiffrin. The results of this work will assist us in building a text generation system for CIRCSIM-Tutor v. 3 which will mimic some of the natural qualities of the speech of human tutors in a simple and efficient manner.