Articles written in Sadhana
Volume 45 All articles Published: 30 July 2020 Article ID 0191
In this paper, we propose an alternate technique to improve the performance of the statistical machine translation (SMT) system. Here, the phrases are re-weighted in light of linguistic knowledge as both syntactic and semantic information. Syntactic knowledge helps to increase fluency whereas semantic similarity helps to incorporate semantic meaning, which is required for adequacy of translated sentences. The scores of the phrases from the phrase-table are re-balanced by expanding and diminishing the weights of the correct phrasesand the incorrect phrases, respectively. Additional knowledge in phrase-table helps in improving overall performance of translation quality. In this work, our proposed methodology achieves an impressive accuracy improvement in terms of BLEU, NIST and RIBES in different domain data. We achieve 58.54 BLEU points,0.7759 RIBES points and 9.684 NIST points for product domain catalog.