Enhancing the Quality of Phrase-Table in Statistical Machine Translation for Less-Common and Low-Resource Languages

Abstract

The phrase-table plays an important role in traditional phrase-based statistical machine translation (SMT) system. During translation, a phrase-based SMT system relies heavily on phrase-table to generate outputs. In this paper, we propose two methods for enhancing the quality of phrase-table. The first method is to recompute phrase-table weights by using vector representations similarity. The remaining method is to enrich the phrase-table by integrating new phrase-pairs from an extended dictionary and projections of word vector presentations on the target-language space. Our methods produce an attainment of up to 0.21 and 0.44 BLEU scores on in-domain and cross-domain (Asian Language Treebank - ALT) English - Vietnamese datasets respectively.