Exploring Document-Level Neural Machine Translation for English-Vietnamese

Abstract

In Neural Machine Translation, the Transformer model has proven to be the state-of-the-art in multiple translation tasks. However, as a Seq2seq model, it can not abstract the contextual information when translating a document from one to another language. In the translation process, there are cases where, without the surrounding contextual information from consecutive sentences, an individual sentence causes ambiguity translations. The document-level approach makes the translation much more coherent and fluent by conserving the connectivity between sentences in the whole document to improve the quality of translation and human readability. Recent works show that models that are able to encapsulate these contextual information gain better results and evaluation than conventional sentence-level models. This paper conducts experiments and analyzes various context-aware models specifically in English-Vietnamese translation tasks.