Measuring Sentences Similarity Based on Discourse Representation Structure

keywords: Sentence similarity, discourse representation structure, structural similarity
The problem of measuring similarity between sentences is crucial for many applications in Natural Language Processing (NLP). Most of the proposed approaches depend on similarity of words in sentences. This research considers semantic relations between words in calculating sentence similarity. This paper uses Discourse Representation Structure (DRS) of natural language sentences to measure similarity. DRS captures the structure and semantic information of sentences. Moreover, the estimation of similarity between sentences depends on semantic coverage of relations of the first sentence in the other sentence. Experiments show that exploiting structural information achieves better results than traditional word-to-word approaches. Moreover, the proposed method outperforms similar approaches on a standard benchmark dataset.
reference: Vol. 39, 2020, No. 3, pp. 464–480