How Custom Glossaries Improve Translation Accuracy by 40%

The Consistency Problem

Imagine translating your product documentation into 10 languages. Without a glossary, the AI might translate "Dashboard" five different ways across your 200-page documentation set. Each translation might be individually correct, but the inconsistency confuses users and makes your product feel unprofessional.

This is not a hypothetical. A 2025 study by Common Sense Advisory found that terminological inconsistency is the number one quality complaint in translated software documentation.

Users reported that inconsistent terminology made them question whether they were reading about the same feature.

What is a Custom Glossary?

A custom glossary is a controlled list of terms and their approved translations in each target language. It tells the translation system: "When you encounter this English term, always use this specific translation."

A glossary entry looks like this:

English	French	German	Spanish

|---------|--------|--------|---------|

Dashboard	Tableau de bord	Dashboard	Panel de control
Workspace	Espace de travail	Arbeitsbereich	Espacio de trabajo
Deploy	Deployer	Deployen	Implementar
API Key	Cle API	API-Schlussel	Clave API

Why Glossaries Improve Accuracy by 40%

The 40% improvement figure comes from a combination of factors:

Factor 1: Eliminates Terminological Ambiguity (15-20% improvement)

Many English terms have multiple valid translations. "Server" in French could be "serveur" (computer) or "serveur" (waiter). "Query" could be "requete" (database query) or "question" (general inquiry). A glossary removes this ambiguity entirely.

Without a glossary, the AI guesses correctly about 60-70% of the time for ambiguous terms. With a glossary, it is correct 100% of the time for defined terms.

Factor 2: Enforces Brand Consistency (10-15% improvement)

Brand names, product names, and feature names should generally not be translated, or should be translated in a specific way. Without a glossary:

• "CloudArchitect" might be translated to the local equivalent of "Cloud Architect" in some languages
• Feature names like "Quick Deploy" might be translated differently by different AI runs
• Technical terms that your audience expects in English might be unnecessarily translated

Factor 3: Domain Accuracy (10-15% improvement)

Domain-specific terms need domain-specific translations. In the medical field, "culture" should translate to the laboratory test term, not the sociological concept. In software, "build" should translate to the compilation process, not the construction process.

A glossary encodes this domain knowledge explicitly.

The Compounding Effect

These improvements compound. A document with 50 glossary-relevant terms appearing an average of 3 times each has 150 translation decisions that are now guaranteed correct. In a 5,000-word document, that represents 150 out of roughly 400 translation decisions (many words translate unambiguously), or about 37.5% of the non-trivial translation decisions.

Building an Effective Glossary

Step 1: Identify Your Terms

Start with these categories:

Product-specific terms: Feature names, product names, UI element labels. These are your highest-priority glossary entries because they appear most frequently and inconsistency is most visible.

Example entries for a project management tool:

• Sprint, Backlog, Story Points, Kanban Board, Burndown Chart

Technical terms: Industry jargon that has specific, non-obvious translations.

Example entries for a cloud platform:

• Instance, Container, Serverless, Load Balancer, Auto-scaling, Deployment Pipeline

Brand terms: Company name, product names, and marketing phrases that should remain untranslated or have specific approved translations.

Example entries:

• TranslateAI (do not translate)
• "Ship Faster" (keep in English or use approved local tagline)

Action terms: Verbs used in your UI that need consistent translation.

Example entries:

• Deploy, Configure, Monitor, Archive, Restore, Export

Step 2: Determine Translations

For each term, determine the correct translation in each target language. Sources for correct translations include:

• Existing translated materials: If you have previously translated content, use the same terms
• Industry standards: Many industries have standardized terminology in multiple languages
• Native speakers on your team: Engineers or support staff who speak the target language
• Competitor analysis: How do competitors translate similar terms?
• Official documentation: Microsoft, Apple, and Google publish style guides with translated terminology for dozens of languages

Step 3: Handle Special Cases

Some terms need special handling:

Do not translate: Brand names, product names, and technical terms that your audience expects in English (API, SDK, URL, CSS).

Transliterate: Terms that should be adapted phonetically rather than translated semantically. This is common for proper nouns in languages with non-Latin scripts.

Adapt, do not translate: Some concepts need cultural adaptation rather than direct translation. A U.S. tax term might need to be replaced with the equivalent local tax concept, not translated literally.

Step 4: Organize and Format

Structure your glossary for easy maintenance:

``csv term_en,term_fr,term_de,term_es,term_ja,notes,category Dashboard,Tableau de bord,Dashboard,Panel de control,ダッシュボード,,UI Deploy,Deployer,Deployen,Implementar,デプロイ,,Action API Key,Cle API,API-Schlussel,Clave API,APIキー,Do not translate 'API',Technical ``

Include notes and categories to help translators and future glossary maintainers understand the reasoning behind each entry.

Glossary Maintenance

A glossary is a living document. It needs regular updates:

When you add new features: New features introduce new terms that need glossary entries.

When you receive translation feedback: Users or reviewers may suggest better translations for existing terms.

When industry terminology evolves: Technical terminology changes over time. "Serverless" did not exist ten years ago.

Quarterly review: Schedule a quarterly review of your glossary with native speakers to catch any entries that feel outdated or unnatural.

Common Mistakes to Avoid

Over-Translating

Not every word needs a glossary entry. Common words like "the," "and," "user," and "data" translate unambiguously and do not need glossary control. Focus on terms that are ambiguous, brand-specific, or domain-specific.

Under-Maintaining

A glossary that was created two years ago and never updated becomes a liability. Outdated entries produce translations that feel stale or incorrect. Assign ownership and schedule regular reviews.

Ignoring Context

Some terms need different translations in different contexts. "View" as a noun (a database view) translates differently than "View" as a verb (view this document). Your glossary should handle these cases, either with separate entries or context notes.

Skipping Validation

Before deploying glossary-driven translations, have a native speaker review a sample of the output. Even correct glossary entries can produce awkward sentences if the grammatical integration is not handled well.

Measuring Glossary Impact

Track these metrics to measure your glossary's effectiveness:

• Consistency score: Percentage of glossary terms that are translated consistently across all documents
• Human reviewer changes: How many terms do human reviewers change? Fewer changes means better glossary coverage
• User feedback: Track translation-related support tickets and feedback across languages
• Translation speed: Glossary-assisted translation is faster because the AI makes fewer uncertain decisions

Get Started with TranslateAI Glossaries

TranslateAI supports custom glossaries that you can upload as CSV files or build directly in the interface. Import your existing terminology, add new entries as you go, and see the immediate impact on translation consistency and accuracy.

Upload your first glossary and translate a document you have previously translated without one. The difference in consistency and accuracy is immediately visible.