Zhaopeng Tu
Noah's Ark Lab, Huawei Technologies, Hong Kong
Yang Liu
Department of Computer Science and Technology, Tsinghua University, Beijing
Zhengdong Lu
Noah's Ark Lab, Huawei Technologies, Hong Kong
Xiaohua Liu
Noah's Ark Lab, Huawei Technologies, Hong Kong
Hang Li
Noah's Ark Lab, Huawei Technologies, Hong Kong
Abstract
In neural machine translation (NMT), generation of a target word depends on both source and target contexts. We find that source contexts have a direct impact on the adequacy of a translation while target contexts affect the fluency. Intuitively, generation of a content word should rely more on the source context and generation of a functional word should rely more on the target context. Due to the lack of effective control over the influence from source and target contexts, conventional NMT tends to yield fluent but inadequate translations. To address this problem, we propose context gates which dynamically control the ratios at which source and target contexts contribute to the generation of target words. In this way, we can enhance both the adequacy and fluency of NMT with more careful control of the information flow from contexts. Experiments show that our approach significantly improves upon a standard attention-based NMT system by +2.3 BLEU points.
PDF (presented at ACL 2017)