Entailment analysis for improving Chinese textual entailment system

Abstract

Textual Entailment (TE) is a critical issue in natural language processing (NLP); many NLP applications can be benefited from the recognition of textual entailment (RTE). In this paper we report our observation on how to improve the Chinese textual entailment system and the experiment results on the NTCIR-10 RITE-2 dataset. To complement the traditional machine learning approach, which treat every input pair equally with the same features and the same process, our system classify different entailment cases and treat them separately. The experiment results show great improvement.