Define your terms once. Enforce them across every language, every time.
Without enforced terminology, AI translation makes independent decisions about how to render your brand's vocabulary. "Dashboard" becomes "Armaturenbrett" in German. "Chargeback" becomes three different words across your app. Glossary Enforcement loads your approved terms as constraints before translation begins — not as a post-hoc check, but as a pre-generation specification.
Glossary Enforcement in Flixu loads your approved terminology as hard constraints into every translation request before the language model generates output. Each term you define — with its approved target-language translations and any forbidden alternatives — is enforced across all 22+ supported languages, for every segment, without exception. The model generates translations with those terms already specified, not selected afterward.
What happens without enforced terminology.
Generic AI translation makes probabilistic decisions about vocabulary. It selects the statistically likely word for each term based on its training data — which is the statistical average of language across the internet, not your approved corporate vocabulary.
The consequences compound quickly. A product with a large UI string count translated without glossary enforcement will accumulate terminology variants over time. "Feature" appears as "Funktion" in one part of the app and "Merkmal" in another. "Submit" translates as "Absenden" on one screen and "Bestätigen" on another. "Dashboard" might appear as "Armaturenbrett" — technically a German word for the same concept, but not what your product means.
Users in international markets notice this. Not always consciously, and often not immediately — but terminology inconsistency accumulates into a signal that the product wasn't really built for their market. According to CSA Research, 76% of software buyers prefer products in their native language; the preference is undermined when the native-language version reads as if the localization was done piecemeal.
The fix isn't to review every translation for terminology after the fact. The fix is to specify the terminology before generation begins.
How Glossary Enforcement works.
Define your terms
Each glossary entry includes three components:
| Field | What it contains | Example |
|---|---|---|
| Source term | The term as it appears in the source language | Dashboard |
| Approved translation | The required term in each target language | DE: Dashboard / FR: Tableau de bord |
| Forbidden alternatives | Terms the model must not use | DE: Armaturenbrett, Schalttafel |
You can define as many terms as needed — product names, UI labels, legal vocabulary, clinical terminology, brand-specific phrases. Upload a CSV or add terms directly through the workspace interface.
Glossary loaded before translation
When a translation request runs, your glossary is loaded into the translation payload before the language model sees the source text. The approved terms are specified as payload constraints — the model receives the source content with the terminology already defined, not as a system prompt suggestion it can override, but as a specified parameter it must use.
Semantic enforcement, not string matching
The enforcement isn't a find-and-replace that substitutes terms after translation. The language model generates text with the approved terms already specified as constraints — which means the surrounding grammar is built around the fixed term from the start. In inflected languages like German or Polish, this matters: inserting a fixed term into already-generated grammar often produces awkward constructions. Specifying the term before generation means the conjugation is native.
This also handles paraphrased contexts. If a source string is reformulated between one sprint and the next — "access the Dashboard" becomes "open your Dashboard" — the model still applies the approved term correctly in both formulations, because the constraint is semantic, not exact-string-dependent.
Verification at scoring
The LQA Terminology Consistency dimension checks that approved glossary terms are present and correctly applied in the translated output. A translation that uses a forbidden alternative or omits a required term scores below threshold on this dimension and is routed for human review.
→ LQA scoring in detail: LQA & Quality Assurance
Glossary Enforcement vs. Brand Voice — what's the difference?
These two features solve adjacent but distinct problems:
| Glossary Enforcement | Brand Voice Manager | |
|---|---|---|
| What it controls | Specific terms — exact vocabulary | Overall tone — formality, style, register |
| Precision level | Binary: this term, not that term | Gradient: more formal, less informal |
| Primary use case | Product names, legal vocabulary, UI labels | Campaign tone, formality register, stylistic preferences |
| Example | "Dashboard" → always "Dashboard" in German | "Use casual du register in German" |
| When missing | Terminology variants across the product | Brand voice drift across languages |
Both load before translation begins and work together in the same pipeline. A product can simultaneously require that "Dashboard" stays untranslated in German and that the overall register is casual — both constraints apply to the same translation request.
→ Brand voice configuration: Brand Voice Manager
When glossary enforcement changes the outcome.
B2B SaaS products with a defined UI vocabulary
A SaaS product with 1,000+ UI strings developed over multiple sprints accumulates terminology decisions. Each time a new string is added and translated without enforced glossary, there's a chance a key term appears differently than it did in previous strings. Over time, the same concept appears as multiple German words in different parts of the interface — a user churning ticket waiting to happen.
Enforcing the glossary from the first sprint means the 1,001st string has the same terminology as the first. Teams using pre-translation glossary enforcement typically find terminology inconsistency drops from 15–25% of reviewed strings to under 2% — because the constraint prevented the error rather than catching it after.
→ SaaS UI localization: Flixu for SaaS Teams
Legal and compliance documents
Legal content requires specific terminology that has defined meaning in the jurisdiction. A contract clause that uses "termination" in English must translate to the specific legal term in German that corresponds to contract termination — not a general word for "ending." A glossary entry that specifies the approved legal term for each concept ensures that terminology is applied consistently across every clause, every document, every jurisdiction.
For compliance teams, the audit trail shows which terms were enforced in which segments — providing a documented record that the translation adhered to the approved terminology standard.
→ Legal compliance localization: Legal Compliance
Healthcare and clinical documentation
Clinical terminology requires absolute precision — a term that means one thing in general language may mean something more specific in clinical context. A glossary that defines approved translations for clinical terms ensures that the same clinical concept appears identically across patient-facing documentation, device labeling, and healthcare provider guides, regardless of which document was translated first or by which pipeline.
→ Healthcare localization: Telehealth & Digital Healthcare
Global marketing with protected brand terms
A brand with protected product names, campaign slogans, or branded phrases in English needs those terms treated consistently across all markets. Some terms should remain untranslated; others have officially approved local variants. Glossary Enforcement lets marketing teams specify both — "ProductName" stays "ProductName" in German, but the campaign tagline has an approved German adaptation that must be used verbatim.
→ Marketing team localization: Flixu for Global Marketing Teams
Frequently Asked Questions
How many terms can I add to a glossary?
+
There's no defined cap on glossary size. You can add as many terms as your vocabulary requires — a small SaaS product might have 50–100 terms; a large enterprise with legal, clinical, and product vocabulary might have thousands. All terms are loaded for each translation request associated with that client profile.
Can I specify different approved translations for the same term in different languages?
+
Yes. Each glossary entry can have separate approved translations per target language. "Chargeback" might be "Rückbuchung" in German and "rétrofacturation" in French — both specified in the same entry. You can also mark a term as "keep in original language" for cases where the English term should remain untranslated in a specific target language.
What happens if a source string contains a term not in the glossary?
+
Terms not in the glossary are handled by the context layer — domain detection, formality, and translation memory inform how the model renders them. Adding a term to the glossary doesn't prevent the rest of the string from being translated normally; it only specifies the constraint for that specific term.
How does enforcement work in languages with complex inflection?
+
Because the glossary constraint is specified before the language model generates text, the model builds the surrounding grammar around the fixed term from the start — it doesn't insert the term into already-generated grammar. This is why the output reads naturally even in inflected languages like German, Russian, or Polish, where inserting a fixed word into an existing sentence often produces awkward constructions.
Can I mark terms as forbidden to translate (i.e., keep them in the source language)?
+
Yes. Glossary entries support a "do not translate" flag — the term appears in the target language output exactly as it appears in the source. This is common for product names, brand names, and technical identifiers that should remain in English across all markets.
Is there an audit trail showing which terms were enforced in which segments?
+
Yes. The LQA Terminology Consistency dimension checks glossary adherence on every segment. Segments where a glossary term was expected and applied correctly contribute to a clean Terminology Consistency score. Segments where a term was missing or a forbidden alternative was used fail the Terminology Consistency dimension and appear in the review queue — providing a documented record of where enforcement succeeded and where human correction was needed.
How does Flixu's glossary enforcement differ from what other TMS tools offer?
+
Most TMS tools with glossary features apply enforcement through one of two mechanisms: a visual highlight for human translators, or a find-and-replace substitution applied after MT output. Flixu loads the glossary as a payload constraint before the language model generates text — the model sees the approved terms as part of the input specification, not as a correction applied to its output. This means the surrounding grammar is built around the fixed terms from the start, and the enforcement holds even when source strings are paraphrased or reformulated between sprints.
Upload your glossary and run your first enforced translation.
Define your critical terms, upload them to a client profile, and see how the output changes. The terminology enforcement is visible from the first translation run.
Related Features
- Brand Voice Manager — Tone and style enforcement alongside glossary constraints
- Translation Memory — Style references that complement glossary enforcement
- LQA & Quality Assurance — Terminology Consistency dimension that verifies glossary adherence
- Auto-Approval Workflows — LQA-based routing that includes terminology compliance
- Glossary & Terminology Management — Topic guide