Improving Translations by Combining Fuzzy-Match Repair with Automatic Post-Editing

Improving Translations by Combining Fuzzy-Match Repair with Automatic Post-Editing John E. Ortega Universitat d’Alacant E-03071, Alacant, Spain jeo10@alu.ua.es Marco Turchi Fondazione Bruno Kessler Trento, Italy turchi@fbk.eu Felipe S´anchez-Mart´ınez Universitat d’Alacant E-03071, Alacant, Spain fsanchez@dlsi.ua.es Matteo Negri Fondazione Bruno Kessler Trento, Italy negri@fbk.eu Abstract Two of the more predominant technologies that professional translators have at their disposal for improving productivity are machine translation (MT) and computer- aided translation (CAT) tools based on translation memories (TM). When transla- tors use MT, they can use automatic post- editing (APE) systems to automate part of the post-editing work and get further productivity gains. When they use TM- based CAT tools, productivity may im- prove if they rely on fuzzy-match repair (FMR) methods. In this paper we com- bine FMR and APE: first a FMR proposal is produced from the translation unit pro- posed by the TM, then this proposal is fur- ther improved by an APE system specially tuned for this purpose. Experiments con- ducted on the translation of English texts into German show that, with the two com- bined technologies, the quality of the trans- lations improves up to 23% compared to a pure MT system. The improvement over a pure FMR system is of 16%, showing the effectiveness of our joint solution. 1 Introduction In recent times, research has shown that translators can be more productive when applying state-of- the-art post-editing techniques (Isabel, 2017). In many cases, the state-of-the-art techniques are ap- plied to improve translation proposals from a trans- lation memory (TM) or directly produced by a ma- c⃝ 2019 The authors. This article is licensed under a Creative Commons 4.0 licence, no derivative works, attribution, CC- BY-ND. chine translation (MT) system. Post-editing tech- niques can be automated and seamlessly integrated into the typical translation pipeline for productiv- ity gains. Two such techniques: fuzzy-match re- pair (FMR) (Ortega et al., 2016) and automatic post-editing (APE) (Chatterjee et al., 2017) have shown to be effective without the initial interven- tion of the translator by offering a repaired trans- lation proposal from a TM in the case of FMR, and an improved MT output in the case of APE. FMR is an automatic post-editing technique typ- ically used with TM-based computer-aided trans- lation (CAT) tools. In TM-based CAT, the trans- lator is offered a translation proposal that comes from a translation unit (a pair of parallel segments) whose source segment is similar to the segment to be translated. When the source segment in the translation unit and the segment to be translated are not identical, which happens very often, the trans- lation proposal needs to be post-edited in order to create the final translation. FMR aims to provide repaired translation hypotheses to reduce the post- editing effort of the original translation proposals by using another source of bilingual information such as an MT system. Some approaches to FMR, like the one by Koehn and Senellart (2010), heav- ily depend on the specific MT system type being used for repairing. Others, such as the …

Content truncated to 500 words