Document-level Mongolian-Chinese Neural Machine Translation Incorporating Target-Side Data Augmentation and Topic Information

The relative scarcity of document-level parallel datasets and lexical ambiguity exist in the research task of Mongolian-Chinese neural machine translation. To solve these problems, we propose a document-level Mongolian-Chinese neural machine translation method that incorporates target-side data augmentation and source-side topic information. First, the method trains a target-side data augmentation (DA) model to extend the target language; then, the source-side language topic information is extracted using a topic model and fused into the source sentences to be translated to assist the model in training and to improve the translation performance of the model. The experimental results show that the method proposed in this paper improves in terms of BLEU value compared with both the sentence-level baseline model and document-level baseline model.