Modeling Dual Read/Write Paths for Simultaneous Machine Translation

Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics Volume 1: Long Papers, pages 2461 - 2477 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Modeling Dual Read/Write Paths for Simultaneous Machine Translation Shaolei Zhang 1,2, Yang Feng 1,2∗ 1Key Laboratory of Intelligent Information Processing Institute of Computing Technology, Chinese Academy of Sciences (ICT/CAS) 2 University of Chinese Academy of Sciences, Beijing, China {zhangshaolei20z, fengyang}@ict.ac.cn Abstract Simultaneous machine translation (SiMT) out- puts translation while reading source sentence and hence requires a policy to decide whether to wait for the next source word (READ) or generate a target word (WRITE), the actions of which form a read/write path. Although the read/write path is essential to SiMT per- formance, no direct supervision is given to the path in the existing methods. In this pa- per, we propose a method of dual-path SiMT which introduces duality constraints to direct the read/write path. According to duality constraints, the read/write path in source-to- target and target-to-source SiMT models can be mapped to each other. As a result, the two SiMT models can be optimized jointly by forc- ing their read/write paths to satisfy the map- ping. Experiments on En↔Vi and De↔En tasks show that our method can outperform strong baselines under all latency. 1 Introduction Simultaneous machine translation (SiMT) (Cho and Esipova, 2016; Gu et al., 2017; Ma et al., 2019; Arivazhagan et al., 2019), which outputs transla- tion while reading source sentence, is important to many live scenarios, such as simultaneous interpre- tation, live broadcast and synchronized subtitles. Different from full-sentence machine translation which waits for the whole source sentence, SiMT has to decide whether to wait for the next source word (i.e., READ action) or translate a target word (i.e., WRITE action) to complete the translation. The sequence of READ and WRITE actions in the translation process form a read/write path, which is key to SiMT performance. Improper read/write path will bring damage to translation performance as compared to the following WRITE actions too many but not necessary READ actions ∗ Corresponding author: Yang Feng. Code is available at: https://github.com/ictnlp/ Dual-Path De: En: Fand (_found it) ich (_I) super (_great) . (_.) I found it great . Alignment Segment pair (a) Segment pairs between the sentence pair. RRWWW|RW|RW RRRWW|RW|RW Source-to-Target (De→En) Target-to-Source (En→De) WRITE Dual READ Alignment Read/Write Path: Fand (_found it) ich (_I) super (_great) . (_.) I found it great . Fand (_found it) ich (_I) super (_great). (_.) I found it great . (b) The duality between the read/write paths in two directions. Figure 1: An example of duality constraints. With dual- ity constraints, the read/write paths of source-to-target and target-to-source translation should project to the same segment pairs between two languages. will result in high translation latency while too few but not sufficient READ actions will exclude indispensable source information. Therefore, an ideal read/write path is that the READ actions com- pared to the following WRITE actions are just suffi- cient and necessary, which means the …

Content truncated to 500 words