Learning to Copy for Automatic Post-Editing

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 6122–6132, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 6122 Learning to Copy for Automatic Post-Editing Xuancheng Huang†, Yang Liu†‡∗, Huanbo Luan†, Jingfang Xu§ and Maosong Sun† †Institute for Artificial Intelligence State Key Laboratory of Intelligent Technology and Systems Department of Computer Science and Technology, Tsinghua University, Beijing, China Beijing National Research Center for Information Science and Technology §Sogou Inc., Beijing, China ‡Beijing Advanced Innovation Center for Language Resources hxc17@mails.tsinghua.edu.cn, liuyang2011@tsinghua.edu.cn, luanhuanbo@gmail.com, xujingfang@sogou-inc.com, sms@tsinghua.edu.cn Abstract Automatic post-editing (APE), which aims to correct errors in the output of machine trans- lation systems in a post-processing step, is an important task in natural language processing. While recent work has achieved considerable performance gains by using neural networks, how to model the copying mechanism for APE remains a challenge. In this work, we pro- pose a new method for modeling copying for APE. To better identify translation errors, our method learns the representations of source sentences and system outputs in an interac- tive way. These representations are used to explicitly indicate which words in the system outputs should be copied, which is useful to help CopyNet (Gu et al., 2016) better generate post-edited translations. Experiments on the datasets of the WMT 2016-2017 APE shared tasks show that our approach outperforms all best published results. 1 1 Introduction Automatic post-editing (APE) is an important nat- ural language processing (NLP) task that aims to automatically correct errors made by machine translation systems (Knight and Chander, 1994). It can be considered as an efficient way to modify translations to a specific domain or to incorporate additional information into translations rather than translating from scratch (McKeown et al., 2012; Chatterjee et al., 2015, 2018). Approaches to APE can be roughly divided into two broad categories: statistical and neural ap- proaches. While early efforts focused on statis- tical approaches relying on manual feature en- gineering (Simard et al., 2007; B´echara et al., 2011), neural network based approaches capa- ble of learning representations from data have ∗Corresponding author: Yang Liu 1The source code is available at https://github. com/THUNLP-MT/L2Copy4APE src I ate a cake yesterday mt Ich esse einen Hamburger pe Ich hatte gestern einen Kuchen gegessen Table 1: Example of automatic post-editing (APE). Given a source sentence (src) and a machine trans- lation (mt), the goal of APE is to post-edit the erro- neous translation to obtain a correct translation (pe). Our work aims to explicitly model how to copy words from mt to pe (highlighted in bold), which is a com- mon phenomenon in APE. shown remarkable superiority over their statisti- cal counterparts (Varis and Bojar, 2017; Chatter- jee et al., 2017; Junczys-Dowmunt and Grund- kiewicz, 2017; Unanue et al., 2018). Most of them cast APE as a multi-source sequence-to-sequence learning problem (Zoph and Knight, 2016): given a source sentence (src) and a machine translation (mt), APE outputs a post-edited translation (pe). …

Content truncated to 500 words