Combining transformation and classification for recognizing textual entailment

Abstract

This paper introduces an approach combining transformation and classification methods for recognizing textual entailment. In transformation model, directional and undirected inference relations are recognized, and text fragments having such relations in text are replaced by the counterparts in hypothesis. In classification model, a hybrid kernel-based approach is introduced, and three kinds of features are employed for classifying entailment. Experimental results show that the combination approach achieves a better performance in comparison with the single classification system.