GWLAN: General Word-Level AutocompletioN for Computer-Aided Translation

Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing, pages 4792–4802 August 1–6, 2021. ©2021 Association for Computational Linguistics 4792 GWLAN: General Word-Level AutocompletioN for Computer-Aided Translation Huayang Li Lemao Liu Guoping Huang Shuming Shi Tencent AI Lab {alanili,redmondliu,donkeyhuang,shumingshi}@tencent.com Abstract Computer-aided translation (CAT), the use of software to assist a human translator in the translation process, has been proven to be use- ful in enhancing the productivity of human translators. Autocompletion, which suggests translation results according to the text pieces provided by human translators, is a core func- tion of CAT. There are two limitations in pre- vious research in this line. First, most research works on this topic focus on sentence-level au- tocompletion (i.e., generating the whole trans- lation as a sentence based on human input), but word-level autocompletion is under-explored so far. Second, almost no public benchmarks are available for the autocompletion task of CAT. This might be among the reasons why research progress in CAT is much slower com- pared to automatic MT. In this paper, we pro- pose the task of general word-level autocom- pletion (GWLAN) from a real-world CAT sce- nario, and construct the first public bench- mark1 to facilitate research in this topic. In addition, we propose an effective method for GWLAN and compare it with several strong baselines. Experiments demonstrate that our proposed method can give significantly more accurate predictions than the baseline methods on our benchmark datasets. 1 Introduction Machine translation (MT) has witnessed great ad- vancements with the emergence of neural machine translation (NMT) (Sutskever et al., 2014; Bah- danau et al., 2015; Wu et al., 2016; Gehring et al., 2017; Vaswani et al., 2017), which is able to produce much higher quality translation results than statistical machine translation (SMT) mod- els (Koehn et al., 2003; Chiang, 2005; Koehn, 1The information of benchmark datasets is in https: //github.com/ghrua/gwlan We asked two sp Wir haben die Meinung von zwei Fachärzten eingeholt. We asked two experts for their opinion. We sp their opinion. a b c 1 specialists 2 specific 3 split Source Sentence Translation 1 specialists 2 specific 3 split Figure 1: Illustration of Different Autocompletion Methods. The translation context is in red. Sub-figure in (a) is the sentence-level autocompletion, where the gray part is the completion generated by MT system. Both (b) and (c) are word-level autocompletion, under- lined text “sp” is the human typed characters and the words in the rounded rectangles are word-level auto- completion candidates. 2009). In spite of this, MT systems cannot re- place human translators, especially in the scenar- ios with rigorous translation quality requirements (e.g., translating product manuals, patent docu- ments, government policies, and other official doc- uments). Therefore, how to leverage the pros of MT systems to help human translators, namely, Computer-aided translation (CAT), attracts the at- tention of researchers (Barrachina et al., 2009; Green et al., 2014; Knowles and Koehn, 2016; Santy et al., 2019). Among all CAT technologies (such …

Content truncated to 500 words