Paraphrase Identification of Marathi Sentences

Abstract

Paraphrasing is the expressing the already said sentence with rewording but without changing its inherent meaning. Hence, different structured sentences may carry the similar meaning and can be identified by paraphrase identification. Paraphrase identification of Marathi sentences is based on the study of structural and semantic analysis of Marathi sentences. Statistical similarity includes similarity on basis of word-set, word-order, word-vector and sumo metric. Whereas, by comparing UNL graphs of two sentences, semantic similarity is calculated which yield semantic equivalence of two sentences. The overall similarity of Marathi sentences is calculated by equally combining the scores of both the statistical and semantic similarity measures. Paraphrase Identification has its important contribution to various NLP tasks like Plagiarism Detection, Text summarization, Question Answering, information Retrieval, Text Simplification and paraphrase detection on SMS.