Application of Unsupervised NMT Technique to Japanese--Chinese Machine Translation

Abstract

Neural machine translation (NMT) often suffers in low-resource scenarios where sufficiently large-scale parallel corpora cannot be obtained. Therefore, a recent line of unsupervised NMT models based on monolingual corpus is emerging. In this work, we perform three sets of experiments that analyze the application of unsupervised NMT model in Japanese--Chinese machine translation. We report 30.13 BLEU points for ZH--JA and 23.42 BLEU points for JA--ZH.