Statistical Machine Translation
Findings of the WMT25 Terminology Translation Task
Pages
23
Time to read
75 mins
Publication
Language
English
Pages
23
Time to read
75 mins
Publication
Language
English
This document is a report on the findings of the WMT25 Terminology Translation Task, which focuses on the translation of specialized terms in high-stakes domains. The task features two tracks: sentence-level translation in the information technology domain and document-level translation in the finance domain. Participants were challenged to translate texts under three modes: without terminology, with proper terminology, and with random terminology. The evaluation criteria included overall quality, terminology accuracy, and consistency. The report details the participation of 13 teams and 20 systems in Track 1 and 4 teams in Track 2. Key findings indicate that the use of proper terminology consistently improves translation quality and accuracy, while random terminology yields smaller gains. The document also highlights the need for improved metrics for document-level translation, as current benchmarks are saturated at the sentence level. The report emphasizes the importance of terminology translation in professional fields such as finance, medicine, and law.