Dynamic Terminology Integration for COVID-19 and Other Emerging Domains

Proceedings of the Sixth Conference on Machine Translation (WMT), pages 821–827 November 10–11, 2021. ©2021 Association for Computational Linguistics 821 Dynamic Terminology Integration for COVID-19 and other Emerging Domains Toms Bergmanis†‡ and M¯arcis Pinnis†‡ †Tilde / Vien¯ıbas gatve 75A, Riga, Latvia ‡Faculty of Computing, University of Latvia / Rain, a bulv. 19, Riga, Latvia {firstname.lastname}@tilde.lv Abstract The majority of language domains require pru- dent use of terminology to ensure clarity and adequacy of information conveyed. While the correct use of terminology for some languages and domains can be achieved by adapting general-purpose MT systems on large volumes of in-domain parallel data, such quantities of domain-specific data are seldom available for less-resourced languages and niche domains. Furthermore, as exemplified by COVID-19 re- cently, no domain-specific parallel data is read- ily available for emerging domains. However, the gravity of this recent calamity created a high demand for reliable translation of critical information regarding pandemic and infection prevention. This work is part of WMT2021 Shared Task: Machine Translation using Ter- minologies, where we describe Tilde MT sys- tems that are capable of dynamic terminology integration at the time of translation. Our sys- tems achieve up to 94% COVID-19 term use accuracy on the test set of the EN-FR lan- guage pair without having access to any form of in-domain information during system train- ing. We conclude our work with a broader dis- cussion considering the Shared Task itself and terminology translation in MT. 1 Introduction This work is part of WMT2021 Shared Task: Ma- chine Translation using Terminologies, which is concerned with improving machine translation (MT) accuracy and consistency on newly devel- oped domains by utilising word and phrase-level terms. We describe Tilde MT systems that are ca- pable of dynamic terminology integration at infer- ence time. Our submissions consist of translations by terminology-enabled general-purpose MT sys- tems for EN-RU, EN-FR, and CS-DE translation *Both authors have contributed equally. directions. Our systems are deliberately trained without consideration for the test domain to follow the spirit of the Shared Task—MT for emerging domains. Despite term collections being noisy, our MT systems with dynamic terminology inte- gration improve term translation accuracy proving their usefulness in dynamic adaptation for novel domains, where training-time domain adaptation methods are not feasible. The remainder of this work describes the meth- ods used for dynamic terminology integration (Sec- tion 2) by describing the tasks of terminology fil- tering, term recognition, and dynamic terminology integration in the translation process. The bulk of Section 2 describes problems due to the low-quality term collections and terminology mismanagement and our solutions to them. We hope that the ex- amples provided will not only illustrate the self- imposed problems by the Shared Task but also will motive reconsidering the purpose and the desired qualities of a term collection in the context of MT. We then briefly describe the experimental setting and results in Section 3 and Section 4 respectively. We conclude our work with a broader discussion considering the Shared Task and terminology trans- lation …

Content truncated to 500 words