Language Understanding and Generation of Complex Tutorial Dialogues

摘要

CIRCSIM-Tutor is an intelligent tutoring system for the domain of cardiovascular physiology, which carries out a natural language dialogue with the user, using a set of tutoring tactics that mimic those employed by expert human tutors. The system has been used extensively by students at Rush Medical College, with positive learning outcomes. The system manages to understand and respond to more than 98% of student inputs using advanced methods of spelling correction and information extraction. Studies of human tutoring sessions have revealed the frequency and variety of hints, and the way they are intertwined with acknowledgments. SGML markup and machine learning have allowed us to classify student initiatives and tutor responses and generate rules for deploying hints and other sophisticated strategies. Comparison of novice and expert tutors has revealed striking differences.