Exploiting Lexical Measures and a Semantic LR to Tackle Textual Entailment in Italian

This paper discusses the participation of the University of Alicante and the Istituto di Linguistica Computazionale in the textual entailment exercise at EVALITA 2009. We present a system based on our previous experiences on the RTE Challenges. The system uses a machine learning classifier fed by features derived from lexical distances, part-ofspeech information and semantic knowledge from SIMPLE-CLIPS, an Italian Language Resource. Although it was our first attempt in recognising entailment relations in Italian and the system was not thought in principle to deal with them, the results achieved encourage us to carry on doing research on this area. We obtain 58% accuracy when applying only lexical features. By considering also semantic knowledge derived from a Language Resource, accuracy reaches 64%.