Adaptation of Back-translation to Automatic Post-Editing for Synthetic Data Generation

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 3685–3691 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 3685 Adaptation of Back-translation to Automatic Post-Editing for Synthetic Data Generation WonKee Lee†∗ Baikjin Jung†∗ Jaehun Shin† Jong-Hyeok Lee†‡ †Departmet of Computer Science and Engineering, ‡Graduate School of Artificial Intelligence, Pohang University of Science and Technology (POSTECH), Republic of Korea {wklee, bjjung, jaehun.shin, jhlee}@postech.ac.kr Abstract Automatic Post-Editing (APE) aims to correct errors in the output of a given machine trans- lation (MT) system. Although data-driven ap- proaches have become prevalent also in the APE task as in many other NLP tasks, there has been a lack of qualified training data due to the high cost of manual construction. eSCAPE, a synthetic APE corpus, has been widely used to alleviate the data scarcity, but it might not address genuine APE corpora’s char- acteristic that the post-edited sentence should be a minimally edited revision of the given MT output. Therefore, we propose two new methods of synthesizing additional MT out- puts by adapting back-translation to the APE task, obtaining robust enlargements of the ex- isting synthetic APE training dataset1. Experi- mental results on the WMT English-German APE benchmarks demonstrate that our en- larged datasets are effective in improving APE performance. 1 Introduction Automatic Post-Editing (APE) seeks to automati- cally correct errors included in the output of a black- box machine translation (MT) system to improve the final translation quality, thereby reducing the effort required for manual post-editing (Allen and Hogan, 2000; Chatterjee et al., 2015; Bojar et al., 2016; Chatterjee et al., 2018). In general, APE can be considered as a task of sequence-to-sequence supervised learning, which requires a considerable amount of human-annotated data. However, con- structing an APE corpus—a set of triplets (Table 1), each of which includes a source text (src), a machine-translated text (mt), and a manually post- edited text (pe)—is labor-intensive work because ∗ Equal contribution to this work. 1Our synthetic APE data is available at https:// github.com/wonkeelee/APE-backtranslation. git src Manipulates the shape of an item . mt Bearbeitet die Form eines Elements an . pe Ver¨andert die Form eines Elements . Table 1: An example of APE triplets from the WMT dataset (Bojar et al., 2017). Boldface words are either incorrect words in mt or post-edited words in pe. post-editors should create pe in principle by mini- mally editing mt while preserving the meaning of src. In fact, the sizes of currently available ‘gen- uine’ APE corpora provided by WMT (Bojar et al., 2016, 2017; Chatterjee et al., 2018, 2019, 2020) are too small to train deep APE models effectively. To overcome the lack of genuine APE corpora, several previous studies have proposed methods to construct synthetic training datasets (Junczys- Dowmunt and Grundkiewicz, 2016; Negri et al., 2018; Lee et al., 2020), and they appear to be par- tially helpful in mitigating the data scarcity prob- lem. One such study is eSCAPE (Negri et al., 2018), which has been shown …

Content truncated to 500 words