A Data Augmentation Method for English-Vietnamese Neural Machine Translation

The translation quality of machine systems depends on the parallel corpus used for training, in particular quantity and corpus. However, building a high-quality large-scale is complex expensive, particularly specific domain Therefore, data augmentation techniques are widely translation. input back-translation method monolingual text, which available from many sources, therefore this can be easily effectively implemented to generate synthetic data. In practice, texts collected different sources websites often have errors grammar spelling, sentence mismatch or freestyle. output reduced, leading low-quality generated by back-translation. study, we propose improving Moreover, supplemented pruning table. We experimented with an English-Vietnamese neural using IWSLT2015 dataset training testing legal domain. results showed that proposed augment translation, thereby quality. our experimental cases, BLEU score increased 16.37 points compared baseline system.