Discourse Understanding and World Knowledge

概要

論文の詳細を見る
Discourse processing is a sequence of dynamic operations on transient information states. As discourse unfolds, the participants keep track of a shared information state that is dynamically updated in a chain of transitions. This shared information state is a bundle of both explicitly asserted and implicitly assumed pieces of information. The role of a linguistic utterance is to make explicit only what does not follow from the currently shared information. Discourse understanding thus combines both linguistic interpretation and commonsense reasoning. It is essential that a theory of discourse account for their interactions.These points are illustrated with an analysis of a piece of discourse-the AI puzzle of 'Missionaries and Cannibals'. The approach is an information-based dynamic discourse analysis with the following features: (1) an information state consists of salient situation types and information pieces, (2) the use of each referring expression type has a set of dynamic constraints in terms of pre- and post-conditions on the information state. The approach exposes the systematic dynamicity in linguistic processing that has tended to be either taken for granted without explanation or neglected in the studies of commonsense reasoning in AI. The same approach is used to analyze equivalent texts in English and Japanese demonstrating its cross-linguistic applicability.
一般社団法人情報処理学会の論文
1992-11-30