DLSITE-2: semantic similarity based on syntactic dependency trees applied to textual entailment

Abstract

In this paper we attempt to deduce textual entailment based on syntactic dependency trees of a given text-hypothesis pair. The goals of this project are to provide an accurate and fast system, which we have called DLSITE-2, that can be applied in software systems that require a near-realtime interaction with the user. To accomplish this we use MINIPAR to parse the phrases and construct their corresponding trees. Later on we apply syntactic-based techniques to calculate the semantic similarity between text and hypothesis. To measure our method’s precision we used the test text corpus set from Second PASCAL Recognising Textual Entailment Challenge (RTE-2), obtaining an accuracy rate of 60.75%.