Detecting and Translating Dropped Pronouns in Neural Machine Translation

Pronouns are commonly omitted in Chinese as well as other pro-drop languages, which causes a significant challenge to neural machine translation (NMT) between pro-drop and non-pro-drop languages. In this work, we propose a method to both automatically detect the dropped pronouns (DPs) and recover their translation equivalences rather than their original forms in source sentences. The detection and recovery are simultaneously performed as a sequence labeling task on source sentences. The recovered translation equivalences of DPs are incorporated into NMT as external lexical knowledge via a tagging mechanism. Experimental results on a large-scale Chinese-English dialogue translation corpus demonstrate that the proposed method is able to achieve a significant improvement over a strong baseline and is better than the method of recovering the original forms of DPs.