Textual Entailment at EVALITA 2009

The rst shared task for recognising textual entailment for Italian was organised during the EVALITA 2009 campaign. The task def- inition followed that of its English counterpart, and consists of the task of when given a pair of texts, determining whether the rst entails the second. A corpus of 800 examples pairs (400 for development/training and 400 for testing) was constructed on the basis of Wikipedia revision histories. Two groups of researchers participated, submitting eight runs in total, with a variation in performance ranging from 0.50 (conform a baseline system that always predicts entailment) to 0.71 accuracy.