Similarity measures based on sentence semantic structure for recognizing paraphrase and entailment

The similarity measure on the sentence level plays an increasingly important role in many applications about text-related areas and natural language processing. In this paper, we employ sentence semantic structures to overcome the difficulty from the variability of natural language expression. We represent a sentence as verb-argument pairs of semantic structures. The similarity between sentences is reflected through the relation between verb-argument pairs. We evaluate the proposed measure on two applications: recognizing paraphrases and entailments. The experimental results show that our method outperforms existing methods in the task of identifying similar sentences.