Combining lexical resources with Tree Edit Distance for recognizing Textual Entailment

Abstract

This paper addresses Textual Entailment (i.e. recognizing that the meaning of a text entails the meaning of another text) using a Tree Edit Distance algorithm between the syntactic trees of the two texts. A key aspect of the approach is the estimation of the cost for the editing operations (i.e. Insertion, Deletion, Substitution) among words. The aim of the paper is to compare the contribution of two different lexical resources for recognizing textual entailment: WordNet and a word-similarity database. In both cases we derive entailment rules that, are used by the Tree Edit Distance Algorithm. We carried out a number of experiments over the PASCAL-RTE dataset in order to estimate the contribution of different combinations of the available resources. © Springer-Verlag Berlin Heidelberg 2006.