Recognizing Textual Entailment Using Lexical Similarity

摘要: We describe our participation in the PASCAL-2005 Recognizing Textual Entailment Challenge. Our method is based on calculating ?directed? sentence similarity: checking directed ?semantic? word overlap between text and hypothesis. use frequency-based term weighting combination with two different lexical similarity measures. best run shows 0.55 accuracy test data, although difference runs not significant. found remarkably optimal threshold values for development data. argue that, addition to accuracy, precision recall are valuable measures consider textual entailment