Towards State-of-the-art English-Vietnamese Neural Machine Translation

Machine translation is one of the most challenging topics in natural language processing. The common approaches to machine translation base on either statistical or rule-based methods. Rule-based translation analyzes sentence structures, requires extensive lexicons with morphological, syntactic, semantic information, and large sets of manually created rules. Statistics-based translation faces the challenge of collecting bilingual text corpora, which is particularly difficult for low resource language pairs as English-Vietnamese. This research aims at building state-of-the-art English-Vietnamese machine translation. Our contribution includes: (1) an enormous effort in collecting training dataset, (2) adaptation of current neural machine for English-Vietnamese translation, (3) an experimental result suggested the unnecessary of Vietnamese word segmentation as a common pre-processing step. Our model achieves a highest BLEU score in comparison with other researches.