Learning approaches for recognizing textual entailment and finding contradiction in texts

概要

論文の詳細を見る
This paper will introduce how machine learning methods and shallow semantic parsing can be applied for natural language understanding. One of the tasks in NLU is recognizing textual entailment, which is to decide whether the meaning of a text can be inferred from meaning of other one. In our work, we conduct an empirical study of the RTE task for Japanese, adopting a machine learning-based approach. We analyze the effects of using bilingual features, machine learning algorithms, and the impact of RTE resources on the performance of a RTE system. We also investigate the use of machine translation for the RTE and show that MT can be used to improve the performance of our RTE systems. We achieved promising results when attended the competitions on NTCIR-9 and NCTIR-10. The second task we would like to present in this paper is finding contradiction in texts. This task is difficult in the sense that we need to deeply understand the texts in order to find contradiction. Previous work on finding contradiction in text incorporate information derived from predicate-argument structures as features in a learning framework. In this paper, we would like to use sallow sematic parsing for these tasks using a simple rule-based framework. We discuss that our methods can be used with the learning based approaches for finding contradiction in texts.
2013-09-05