Recognizing Textual Entailment for Italian EDITS @ EVALITA 2009
Elena Cabrio, Yashar Mehdad, Matteo Negri, Milen Kouylekov, Bernardo Magnini · 2009
Abstract. This paper overviews FBK’s participation in the Textual Entailment task at EVALITA 2009. Our runs were obtained through different configurations of EDITS (Edit Distance Textual Entailment Suite), the first freely available open source tool for Recognizing Textual Entailment (RTE). With a 71 % Accuracy, EDITS reported the best score out of the 8 submitted runs. We describe the sources of knowledge that have been used (e.g. extraction of rules from Wikipedia), the different algorithms applied (i.e. Token Edit Distance, Tree Edit Distance), and the Particle Swarm Optimization (PSO) module used to estimate the optimal cost of edit operations in the cost scheme. Two different dependency parsers for the annotation of the data in the preprocessing phase have been compared, to assess the impact of the parser on EDITS performances. Finally, the obtained results and error analysis are discussed.