UAIC Participation at RTE4
Adrian Iftene, Mihai Alex Moruz · 2008
Textual entailment recognition is the task of deciding, when given two text fragments, whether the meaning of one text is entailed from the other text. This year, at our second participation in the RTE competition, we improve the system built for the RTE3 competition. The main idea of our system is to map every word from hypothesis to one or more words from the text. For that, we transform the hypothesis making use of extensive semantic knowledge from sources like DIRT, WordNet, VerbOcean, Wikipedia and Acronyms database. After the mapping process, we associate a local fitness value to every word from hypothesis, which is used to calculate a global fitness value for current fragments of text. The global fitness value is decreased in cases in which a word from hypothesis cannot be map to one word from the text or when we have different forms of negations for mapped verbs. In the end, using thresholds identified in the training step for global fitness values, we decide for every pair from test data if we have entailment or not. 1.