Glossary-aware translation gives a language model the client’s approved terms, relevant translation-memory matches and style rules for each segment, then checks the output against them, so a post-editor is correcting meaning and nuance rather than re-applying the glossary by hand.
All Translation & Localisation →
Most machine translation glossary features match a source term and force a target term. That breaks on inflected languages, where the forced term arrives in the wrong case or gender; on terms that depend on context, where one source word has two approved translations; and on do-not-translate items that should pass through untouched. The result is a segment that follows the glossary and reads wrongly, which a post-editor then has to fix anyway.
Language models handle this better because they can be given the relevant terms with their notes and asked to use them grammatically. But a model that follows instructions most of the time is not the same as one that follows them every time. So we treat the glossary as both an input and a test: the right entries go into each request, and the output is checked afterwards for every term that should be there. A segment that fails the check is retried or flagged, never silently passed.
from XLIFF or the CAT tool
terms, memory matches, style rules
engine chosen for the pair and domain
required terms present, forbidden absent
in Trados, memoQ or Phrase
segments that fail the check
Give it only the glossary entries relevant to each segment, with context notes and forbidden alternatives, rather than the whole glossary, then verify the output: check that each required target term appears in an acceptable form and no forbidden term does. Segments that fail are retried with the failure stated, and flagged to the linguist if they fail again.
That is where language models have an advantage over simple term replacement, because they can inflect an approved term correctly in context. The verification step has to match inflected forms rather than exact strings, which we handle with lemmatisation or explicit form lists per language, tested on your content.
Yes. Jobs can be routed by client or content type: confidential material to models you host yourself, or to providers under no-training, no-retention terms, and less sensitive content wherever quality and cost are best. The routing rules are written down and auditable.
What we built, what stage it is at, and a link to the live thing wherever there is one.
yourOS is an operating system with an AI at its heart. You talk to it like a person, out loud or typed. It makes your calls, plans your days and works on your projects with a crew of agents, on your own private server. Our own product, built from the first commit; the waitlist is open.
Website, backend and app assets for Sonora, a free AI sound-therapy app on iOS and Android published by Cognitive AI.
Brand, website, prototype, apps on five platforms, the AI backend and the algorithm-intelligence pipeline for Podswap, a social growth engine that runs without a human in the loop.
In closed beta: Roxy, a private AI assistant with permanent memory running on infrastructure the user owns. Brand, site, product, backend and conversational design.
Live at coherence.com: brand identity generated with AI agents we directed, the motion system and website for the group behind Sonora, AboutAI, WakeUp and MudShadows. Agent work remains in build.
Live at domainbooth.com for DomainBooth, a premium domain brokerage: a new animated mark, every page redesigned and rebuilt with designed mobile layouts, and a custom CMS for its sales record, featured names, press and enquiries.
We start with one client’s termbase and a sample of real jobs.
30-minute technical call · no deck