AN ABDUCTIVE APPROACH TO DISCOURSE UNDERSTANDING USING TENSE AND OLD-NEW INFORMATION AND ITS APPLICATION TO NI-BASED TROUBLESHOOTERS

1995 Fiscal Year Final Research Report Summary

• Project/Area Number: 05451160
• Research Category: Grant-in-Aid for General Scientific Research (B)
• Allocation Type: Single-year Grants
• Research Field: Intelligent informatics
• Research Institution: SOPHIA UNIVERSITY
• Principal Investigator: ISHIKAWA Akira SOPHIA UNIVERSITY,ENGLISH LANGUAE,PROFESSOR, 外国語学部, 教授 (10138373)
• Project Period (FY): 1993 – 1995
• Keywords: ABDUDTIVE REASONING / NATURAL LANGUAGE INTERFACE / DISCOURSE ANLYSIS / DISCOURSE REPRESENTATION THEORY16FA05 : DISCOURSE REFERENT / LABELLED DEDUCTIVE SYSTEM / ATMS / DIALOG GAMES / 対話ゲーム

Research Abstract

This study has as its main goals the development of a reasoning system which can reflect tense and old-new information of NL discourse and its application to a practical troubleshooting system as a component of its NL interface. Although the study is at an end in terms of its assigned term, it has covered no more than the fundamentls of the workings of NL temporal expressions as viewed from the viewpoints of some of the most promising theories to be used in such a reasoning system. It has been established that in order for a machine to come close enough to NL inferencing to support reasnable man-machine dialogs, it is necessay for the reasoning procedure to take into account the discourse referents constantly introduced into the text and updated as the text unfolds. For management of discourse referents, several proposals have been considered to assess their strengths and weaknesses. The results of this assessment are summarised in our final report. Since discourse referent management involves several different modes of logical reasoning, the language to be used for such purposes should be rich in expressivity and easy to handle, Muskens' lambda-categonical language has received a higher rating in our appraisal in this connection over Kamp's original DRT.It has also been found out that temporal reasoning presupposes a method of extracting temporal information from NL expressions. No reasonable proposal has been made of an algorithmic method with which the temporal meaning of an expression can be compositionally deducible from that of its components. Ter Muellen's idea of relating the aspectual characteristics of phrases to the behavior of generalized quantifiers is favorably considered to support one way of developing the alogorithmic method though it is still in the stagee of a rough sketch. Our original ambition was to present a concrete instance of a algorithmic method treating tense and old-new information as reflected in the Japanese tensee system and the opposition of wa-ga and to turn it into a deductive system. As indicated above, the task remains to be undertaken in a future study.