Using hypernymy acquisition to tackle (part of) textual entailment

Abstract

Within the task of Recognizing Textual Entailment, various existing work has proposed the idea that tackling specific subtypes of entailment could be more productive than taking a generic approach to entailment. In this paper we look at one such subtype, where the entailment involves hypernymy relations, often found in Question Answering tasks. We investigate current work on hypernymy acquisition, and show that adapting one such approach leads to a marked improvement in entailment classification accuracy.