Plan-based dialogue management in a physics tutor
ANLC '00 Proceedings of the sixth conference on Applied natural language processing
Selected AI-related dissertations
ACM SIGART Bulletin
| Hi-index | 0.00 |
THIS THESIS DESCRIBES MENO-TUTOR, A COMPUTER PROGRAM FOR TUTORING. THE PROGRAM ADAPTS ITS UTTERANCES TO THE CONTEXT OF STUDENT AND DISCOURSE HIS- TORY, I.E., IT RESPONDS DIFFERENTLY TO THE KNOWLEDGEABLE STUDENT AND THE CONFUSED ONE. THE PROGRAM USES KNOWLEDGE ABOUT TUTORING STRATEGIES, COMPLEX COMMUNICATION SKILLS, AND ITS ABILITY TO INFER THE LEVEL OF THE STUDENTS'' KNOWLEDGE TO GENERATE REASONABLE TUTORING DISCOURSE. IT CAN QUESTION A STU- DENT ABOUT THE SUBJECT DOMAIN, PROBE HIM FOR POSSIBLE MISCONCEPTIONS, AND CAN CHANGE ITS STRATEGIES IF THE STUDENT IS NOT PROGRESSING WELL. THE PLANNING MECHANISM FOR MENO-TUTOR IS BEST DESCRIBED AS A SET OF DE- CISION-MAKING STATES ORGANIZED INTO THREE LEVELS. EACH STATE PROVIDES A CONSTRAINT OR DECISION ABOUT THE FORM AND CONTENT OF THE UTTERANCE. THE LEVELS SUCCESSIVELY DEFINE THE ACTIONS OF THE NEXT LEVEL AND CONSTRAIN THE PEDAGOGICAL, STRATEGIC AND TACTICAL FORMATION OF THE UTTERANCE. STATES ARE LINKED TO OTHER STATES BY A STRUCTURE WHICH IS NOMINALLY AN OR GRAPH, BUT MAKES TWO NOTABLE DEVIATIONS. THE FIRST IS A SET OF DEFAULT TRANSITIONS THAT REPRESENT CROSSLINKS THROUGH THE STATES AND ARE RESPONSI- BLE FOR THE TRADITIONAL SECTIONS OF DISCOURSE, E.G., INTRODUCTION OF A TOPIC, QUESTIONING THE STUDENT ABOUT THE TOPIC, AND TERMINATING THAT TOPIC. THE SECOND IS A SET OF "META-RULES" WHICH FUNCTIONALLY REPRESENT THE SHIFTS OBSERVED IN CLASSIC HUMAN TUTORING; THEY EXPRESS THE HIGH-LEVEL TRANSITIONS