2025 will be the year of MT Quality Estimation
Market watchers predict the great normalization of AI for 2025. Time to reap the benefits: not just talk the talk, but walk the walk. But what does that mean for language service providers and translation buyers that already use Machine Translation to the max and realize savings of 50% with Post-Editing?
MT Quality Estimation is proving to be a very practical and profitable AI-driven tool to automate the post-MT production process. Where MT models are trained to produce the best possible translation output, MTQE models are trained to find the errors in the MT output. This dedicated training makes them perform amazingly well. See this Real-World Guide to QE Scoring for a whole range of examples where fully automatic QE detects for instance polysemy errors, omissions, number errors, spelling errors, tone differences, contextual sensitivity, technical terminology and word order errors, even subtle translation errors.?
Following the wide adoption of MT in the localization industry, we can expect that LSPs and translation buyers will want to leverage their MT investments further by implementing the complementary automated Quality Estimation technology and realizing the benefits of full control over an automated AI-driven translation process.?
Which other AI tool is out there that can help the translation world now to book dramatic cost savings and boost production volumes?
Logrus Global
2 个月....hopefully, the year when the industry will realize that segment-level quality score is quite meaningless (especially for workflow decisions), and translation quality can only be evaluated on a sample not less than 200 segments that constitutes single unified piece of content.
MA, PGCert, DipTrans (CIOL), BSBA | Localization, Terminology and Digital Marketing Expert | Board Member, TermNet, ELETO | Greek National Expert for ELOT/TC 21 and ISO/TC 37 | Language & AI Researcher
2 个月Jaap I am so eager to test it in the English into Greek language pair for my PhD and work! Happy New Year!
Translator
2 个月Oh no! This is so sad. I can already foresee the copious amounts of unnatural and incorrect translations rising up in 2025 :( Especially from large corporations such as Microsoft - their translations have been on a steady decline quality-wise ever since this AI fad took on.
I'd like to throw in my prediction of "orchestration", which includes a component for quality estimation. We'll have smaller, LLMs (aka SLMs) dedicated to translation, and larger LLMs for fact checking, consistency checking, domain appropriateness and recovery from critical errors. At the end of 2025 we'll have deepened our confidence in automatic translation.
M.A. Holder | Language Development Team Lead | CAT Tool Accessibility Advocate | Localization Section Head | Assistant Lecturer | Localization Business Coach | Mentor | English<>Arabic Translator | Reviewer | LQA|Tester
2 个月I am eager to test it in the English into Arabic language pair. I have tested CometKiwi and it performs well in the Pharmaceutical and HR domains but it didn't perform well in the Gaming domain.