An ILP Formulation of Abductive Inference for Discourse Interpretation

概要

論文の詳細を見る
Abduction is inference to the best hypothesis to explain observations. Hobbs et al.6) demonstrate that abduction gives a reasonable formalization of the process of discourse understanding, and several natural language processing (NLP) tasks can be resolved with a single abduction-based framework. However, there is a critical problem with this approach: the computational cost of abduction. The task of abductive reasoning quickly becomes intractable as the amount of background knowledge is increased to cover the millions of axioms necessary for robust discourse processing. This computational bottleneck is preventing abductive reasoning from benefiting from the recent advances in computational resources for for commonsense reasoning. In this paper, we propose an efficient implementation of Hobbs et al.'s abductive discourse interpretation framework, weighted abduction. Our framework transforms the problem of explanation finding in weighted abduction into a linear programming problem. Our experiments showed that our approach efficiently solved problems of plan recognition and outperforms an existing system for weighted abduction.
2011-09-09