LOC.01
Glossary-Aware Translation
LLM translation given the relevant termbase entries, memory matches and style rules per segment, then checked for every required term, so post-editors correct meaning rather than terms.
Open pageIndustry
AI for translators,
in your terminology.
Fluent is easy. Correct terminology is the work. Glossary-aware translation, quality estimation and terminology management, built into the CAT tools and memories you already run. Fluency is the easy part; the client’s terminology is the work.
In context
AI for translators means machine translation and checking tools that follow a client’s glossary, translation memory and style guide, and that flag their own doubtful segments, so linguists spend their time post-editing and reviewing where it matters instead of everywhere.
Where it breaks today
01
Fluent and wrong
Large language models translate fluently, which makes their errors harder to see: a product name translated that should not be, a term rendered three different ways in one manual, a dropped negation in a safety warning. Fluency hides exactly the mistakes a reviewer most needs to find.
02
Terminology lives everywhere
Client glossaries in spreadsheets, termbases in the CAT tool, decisions buried in query logs and email. Generic AI translation for business ignores all of them, and the linguist re-applies them segment by segment.
03
Review effort is spread evenly
Without a reliable signal for which segments are risky, every segment gets the same attention. Reviewers either check everything, which is slow, or sample, which lets the expensive error through.
OUR APPROACH
AI localisation that respects the linguist’s workflow
We build for language service providers and in-house localisation teams, inside the tools they already use. Trados Studio, memoQ, Phrase and XTM stay the working environment; our systems connect through their APIs and the standard exchange formats (XLIFF, TMX, TBX), so translation memories and termbases remain the source of truth and nothing is locked into a new platform.
Three things decide whether AI translation is worth having. Does it follow the client’s terminology every time, not most of the time? Does it say which segments it is unsure of, so review effort goes where the risk is? And does every correction a linguist makes flow back into the termbase and memory, so the next job starts better? Workflows can be set up to fit ISO 18587 post-editing of machine translation and ISO 17100 processes; whether your process meets those standards remains your determination.
HOW IT DEPLOYS
PHASE 01
Measure on your content
Compare engines and prompts on a sample of real jobs per language pair and domain, scored by your own reviewers, before choosing anything.
PHASE 02
Wire in the terminology
Termbases, do-not-translate lists and style rules enforced at translation time and checked afterwards.
PHASE 03
Route review by risk
Quality estimation sends doubtful segments to full review and lets confident ones go to lighter checks, with sampling to keep it honest.
What we build
01
Glossary-aware translation
LLM translation constrained by the client’s termbase, translation memory matches and style guide, with every term decision visible to the post-editor.
SEE GLOSSARY-AWARE TRANSLATION →02
Translation QA
Quality estimation per segment, plus checks for terminology, numbers, tags, omissions and meaning shifts, so reviewers start with the riskiest segments.
SEE TRANSLATION QA →03
Terminology management
Term extraction from source and reference material, candidate translations for terminologists to approve, and termbases kept current across clients.
SEE TERMINOLOGY MANAGEMENT →04
Model choice and routing
Different engines suit different language pairs and domains. Routing each job to the one that measured best on your content, including models you host yourself.
SEE MODEL ROUTING →Localisation use cases
-
-
LOC.02
Translation QA
Reference-free quality estimation and MQM-style findings per segment, plus coded checks for numbers, tags and terms, calibrated on your reviewed jobs so review starts with the riskiest segments.
Open page -
LOC.03
Terminology Management
Term extraction from source and reference files, evidence of existing translations and drafted termbase entries, synced to the CAT tool. A terminologist approves every entry.
Open page
Questions
Questions about Translation & Localisation.
Will AI replace translators?
It has already changed the work. Much commercial translation now starts from machine output that a linguist post-edits, and that share is growing. What does not go away is accountability for meaning: legal, medical, safety and brand content still needs a qualified person who understands both languages and the consequences of an error. The systems we build are designed to make that person faster, not to remove them.
What is AI localisation?
AI localisation uses machine translation and language models to adapt content for another market, covering not only words but terminology, formats, units, tone and cultural references. Done well, it follows the client’s glossary and style guide, flags uncertain segments for review and feeds corrections back into the translation memory. Done badly, it produces fluent text that is subtly off brand or wrong.
Can AI translation be trusted for business documents?
For internal understanding and low-risk content, often yes. For contracts, regulatory submissions, medical or safety material, and published marketing, raw machine output should not be used without review by a qualified linguist. Quality estimation helps decide which segments need full review, but the decision about what is fit to publish belongs to a person.
Which is better for translation, DeepL or an LLM?
It depends on the language pair, the domain and how much the job relies on terminology and context. Dedicated engines are often strong and consistent at sentence level; large language models handle context, style instructions and glossaries more flexibly. We measure the candidates on a sample of your real content, scored by your reviewers, and route accordingly.
Does it work with Trados, memoQ and Phrase?
Yes, through their APIs or plugins where available and through XLIFF, TMX and TBX files otherwise. The CAT tool stays the linguist’s working environment, and translation memories and termbases stay the record, so adopting AI does not mean moving every client to a new platform.
