Cross-lingual neural fuzzy matching for exploiting target-language monolingual corpora in computer-aided translation

Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 7532 - 7543 December 7-11, 2022 ©2022 Association for Computational Linguistics Cross-lingual neural fuzzy matching for exploiting target-language monolingual corpora in computer-aided translation Miquel Esplà-Gomis, Víctor M. Sánchez-Cartagena Juan Antonio Pérez-Ortiz, Felipe Sánchez-Martínez Dep. de Llenguatges i Sistemes Informàtics Universitat d’Alacant E-03690 Sant Vicent del Raspeig (Spain) {mespla,vmsanchez,japerez,fsanchez}@dlsi.ua.es Abstract Computer-aided translation (CAT) tools based on translation memories (MT) play a promi- nent role in the translation workflow of profes- sional translators. However, the reduced avail- ability of in-domain TMs, as compared to in- domain monolingual corpora, limits its adop- tion for a number of translation tasks. In this paper, we introduce a novel neural approach aimed at overcoming this limitation by exploit- ing not only TMs, but also in-domain target- language (TL) monolingual corpora, and still enabling a similar functionality to that offered by conventional TM-based CAT tools. Our ap- proach relies on cross-lingual sentence embed- dings to retrieve translation proposals from TL monolingual corpora, and on a neural model to estimate their post-editing effort. The paper presents an automatic evaluation of these tech- niques on four language pairs that shows that our approach can successfully exploit mono- lingual texts in a TM-based CAT environment, increasing the amount of useful translation pro- posals, and that our neural model for estimat- ing the post-editing effort enables the combi- nation of translation proposals obtained from monolingual corpora and from TMs in the usual way. A human evaluation performed on a single language pair confirms the results of the automatic evaluation and seems to in- dicate that the translation proposals retrieved with our approach are more useful than what the automatic evaluation shows. 1 Introduction Despite the recent advances in machine translation (MT), translation memories (TM) play a prominent role in the translation workflows of language ser- vice providers, translation departments, and free- lance translators; the 2020 edition of the annual European Language Industry Survey (ELIA et al., 2020) reported TMs to be the most popular re- source among freelance translators and a relevant target of investment for companies and translation departments. Even when the perceived quality of the translation proposals from TM and MT is quite similar, professional translators need more time to post-edit MT outputs than TM proposals (Sánchez- Gijón et al., 2019). In addition, the use of TMs foster terminology consistency and is especially suited to repetitive translation tasks as they allow for convenient recycling of previous translations. Formally, a TM (Isabelle et al., 1993) may be defined as a collection of translation units (TU), that is, source-target sentence pairs (s, t) which are mutual translations. They are conveniently ex- ploited by computer-aided translation (CAT) tools which, for each new sentence to be translated s′, retrieve from the TM the subset of TUs {(s, t)} whose source sentences s are most similar to s′. Translators then choose the TU that better fits their needs and make the appropriate changes to t, the target sentence in the selected …

Content truncated to 500 words