Low Resource Neural Machine Translation

Abstract

Neural machine translation (NMT) is the current state-of-the-art approach for machine translation. However, NMT models should be trained with a large amount of data, making NMT in low-resource scenarios a tricky issue. In this paper, we concluded three categories of methods for low-resource NMT. Firstly, data augmentation is the most direct solution, producing extra parallel corpus. Secondly, multilingual NMT model can improve the performance of low-resource languages. Finally, multimodal NMT is especially useful because multimodal information is easy to acquire online. We also illustrate some promising directions to further explore in the future.