How to Enforce Terminology in Machine Translation Post-Editing

Fifteen pages left to edit, deadline’s tomorrow, and the MT output’s already killed the client three times: “Geschäftsrecht” shows up as “business law”, “commercial justice”, and “commercial law” in the same document. The client sees the inconsistency, you spend 20 minutes manually standardizing, and the final text still reads like it was Frankensteined together.

Sound familiar? This is the cost of loose terminology in MTPE workflows. On average, 20-30% of machine translation errors boil down to a single problem: inconsistent use of key terms. The good news: it’s solvable. The bad news: most agencies ignore it until they lose a major client over quality.

This guide shows you how to set up a system where the machine enforces your terminology automatically, and which tools actually deliver on that promise.

Why MT Breaks Your Glossary (Even When You Have One)¶

The first paradox: you uploaded your glossary to Google Sheets, briefed the team, and the MT engine still renders “UI” as “user interface”, “interface”, and “user-facing interface” in a single document.

Here’s why: a disconnected glossary isn’t enforcement—it’s a suggestion. Google Sheets, Confluence, even Excel in Dropbox—the machine translation engine doesn’t see them. You do. Your post-editor does. The MT system doesn’t. Result: no change in output.

According to research from ACL (Association for Computational Linguistics), 68% of language service providers using glossaries store them in spreadsheets disconnected from their translation engines. Outcome: glossaries reduce errors by 2-3% instead of the hoped-for 20-25%.

The second paradox: even integrated glossaries often make quality worse. Here’s an example: you tell the system “Geschäftsrecht” = “business law”. It applies this rule, but when the phrase reads “Geschäftsrecht breaks the contract”, your output becomes “business law break contract” or worse—grammatically mangled because the MT engine forced a term into a position where it doesn’t fit. The post-editor then spends 5 minutes fixing the grammar, defeating the purpose.

Root cause: neural MT doesn’t understand grammar. If you mandate “business law” everywhere without context, the system gets caught between enforcing your term and maintaining grammatical correctness. On average, 40% of glossary enforcement decisions create trade-offs that hurt more than they help.

How Glossaries Actually Work (and Why Small Beats Large)¶

When a glossary is properly integrated—into memoQ, Trados, or a specialized platform—the flow looks like this:

1. Planning phase: the system reads your glossary and maps it. It sees “API” → “application programming interface” and holds that in memory.
2. Translation phase: as the MT engine processes text, it “remembers” the glossary entries. When it hits “API”, it doesn’t choose between variants—it enforces the glossary term.
3. Post-translation QA: the system flags any glossary violations or inconsistencies before delivery.

In practice, this achieves 15-25% reduction in terminology errors, with one condition: the glossary stays small and precise.

Why small beats large: a 300-term glossary is 300 words you curate and update regularly. They cause frequent errors? They’re critical for the client? They’re in. A 2000-term glossary is “add everything we might ever use.” Result: - The machine tries to force all 2000 terms into every translation - Grammar collapses - Post-editing becomes longer, not shorter

Phrase’s research even recommends starting with 100-150 critical terms. This gives you 80% quality without the grammar destruction risk.

Building a Glossary from Scratch¶

Here’s a plan that works across 15+ projects:

Step 1: Term extraction (30-60 minutes)

Take your source document (English original) and scan for: - Words that repeat (>3 times in the document) - Industry-specific jargon (medical, legal, technical) - Terms the client said “we call it this”

Result: 80-200 candidates.

Step 2: Clean & prioritize (20-30 minutes)

Delete: - Generic words (“the”, “be”, “have”—the engine already knows these) - Synonyms (“customer” and “client”—pick one) - Obscure terms that clutter the glossary

Remaining: 120-180 terms.

Step 3: Translate & define (60-120 minutes)

For each term, add: - Translation to your target language - Brief definition (1-2 sentences) - Example context (one line)

Example:

Step 4: Test (1 document)

Load the glossary into your tool (memoQ, Trados, platform), translate one document, check: - Does the machine apply the terms? - Does grammar survive? - Does your post-editor work faster?

If not, cut the glossary in half and retry.

Step 5: Iterate (after each project)

What terms does MT still mistranslate, glossary or not? Add them next time or remove them if they’re too context-dependent.

Tools: Comparing Approaches¶

Three categories exist. Here’s the breakdown:

Which should you choose? - Solo freelancer (1-2 projects/month): Disconnected glossary + memoQ Free (discipline required). ROI: minimal. - Small agency (5-10 projects/month): CAT tool (memoQ or Trados) with integrated glossary + Translation Memory. ROI: 3-6 months. - Medium/large agency (20+ projects/month): Platform with built-in enforcement + QA validator. ROI: 1-3 months.

Integrating a Glossary into memoQ (The Standard)¶

MemoQ is the de facto standard for MTPE in Europe. Here’s the process:

1. Create a termbases

In memoQ: Workspace → Termbases → New → Bilingual. Name it “Project_Client_2026”.

2. Add terms manually or by import

Manual (for 100-150 terms): - Concept - Term EN - Term Target - Definition - Context

Or via CSV:

Concept,English,TargetLanguage,Definition
Payment Method,Credit Card,[Translation],Method of payment using a credit card

3. Attach to your project

Project settings → Termbases → Assign your termbases. Now memoQ: - Highlights new terms in your document - Suggests the correct translation from the glossary - Allows quick lookup during editing

4. Enable QA

Critical step. The QA module checks: - Are all glossary terms used consistently? - Are there unrecognized term variations?

Without QA, the glossary is a suggestion. With QA, it’s enforcement.

5. Export & reuse

A termbases built for Project A can be reused on Project B. This saves ~80% of setup time on future projects.

Comparing Against Competitors¶

Best Practices: How to Maximize ROI¶

1. Build a master glossary

Don’t restart with each project. Create one master glossary containing every term you’ve ever standardized. New project? You start with 70-80% of terms already defined, add the remaining 20-30%. Time saved: 2-3 hours per project.

2. Version and archive

Glossary v2.1 (Jan 2026) for Project A. Glossary v2.5 (Mar 2026) for Project B. If a client says “keep ‘API’ as-is, don’t translate”, you record that decision. Future team members know exactly why.

3. Calculate ROI

• Time to build glossary: 3-5 hours
• Time saved on first project: 30-60 minutes (post-editor works faster)
• Time saved on second: 45-90 minutes (fewer lookups needed)
• Glossary breaks even: project three

An agency doing 20 projects/year recoup ~600-1200 hours. At $30/hour, that’s $18-36K in savings. Tool cost: ~$500/year. Good investment.

4. Bill the client

When a client says “terminology consistency is critical”, charge for it: - Glossary creation: 2-4 hours = $60-120 - Integration: 1 hour = $30 - Per-project enforcement surcharge: base rate + 10%

Clients understand that quality costs and happily pay.

Common Pitfalls (and How to Avoid Them)¶

Pitfall 1: Glossary without context

You add “Bank” → “Banco” to your glossary. But the document contains “Bank of England”, “bank account”, and “riverbank”. Which should use the glossary term? The MT engine gets confused, grammar breaks.

Fix: Every term needs a definition and example context.

Pitfall 2: Outdated glossary

Months one through six, the glossary updates regularly. Then it stagnates. Six months later, you’ve added 15 new critical terms that aren’t in the glossary.

Fix: Review and update monthly. Archive obsolete terms quarterly.

Pitfall 3: Glossary without QA enforcement

You set up the glossary but don’t enable QA. Post-editors ignore the suggestions. Edit times don’t improve.

Fix: QA must be mandatory—a hard gate before delivery.

Pitfall 4: Bloated glossary

You add 5000 terms. The MT engine tries to force all of them into every output. Quality plummets.

Fix: Start with 100-150 critical terms. Grow only when those terms cause repeated errors.

Manual Consistency Control (When You Can’t Use a Glossary)¶

If you’re stuck with Google Translate or a platform without glossary support, do this:

Post-editing checklist: 1. Note the first occurrence of each key term 2. Use ctrl+F to find every instance in the document 3. Standardize all variants to match the first occurrence

Time cost: 5-10% of edit time, but it works.

QA tools: - Linguistic assistants (Antidote, LanguageTool) offer basic checks - memoQ Free Edition has limited QA (better than nothing)

FAQ¶

Here are answers to questions that come up after reading:

Does a glossary actually improve machine translation quality?

Yes—but only with proper integration. A disconnected glossary (stored in Google Sheets or similar) gets ignored by the MT engine. An integrated glossary is automatically applied. Real-world result: 15-25% reduction in terminology errors when the glossary contains 100-200 critical terms.

How many terms should go into a glossary?

Start with 100-150 critical terms—the ones that cause the most errors or matter most to your client. Oversized glossaries (1000+ terms) often degrade quality because the MT engine can’t fit all those terms grammatically into your output. A lean, well-curated glossary beats a bloated one.

Can I use glossaries with DeepL or Google Translate?

DeepL supports glossaries in limited language pairs (English, French, German, Italian, Spanish, Japanese). Google Translate offers no native glossary feature. For complete control, you need CAT tools (memoQ, Trados) or specialized platforms.

How often should I update the glossary?

Review errors from completed projects monthly and add new terms. Archive outdated terms quarterly. A glossary without updates loses relevance within six months.

Will a glossary eliminate the need for project management or CMS?

No. A glossary + TM + QA handles quality control; it doesn’t replace project management. You still coordinate people, manage deadlines, and track deliverables. Tools automate repetitive quality tasks, nothing more.

We’re a small team (2-3 translators). Is a glossary worth setting up?

Absolutely. Even small agencies gain consistency across projects, easier training for new hires, and measurably higher quality. ROI happens faster due to lower overhead.

How ChatsControl Can Help¶

If you’re looking for an MTPE tool with built-in terminology enforcement, ChatsControl delivers:

• Translation brief with terminology: Before translating, you can specify preferred terms, style, and context. The AI applies this across all translation stages (plan → translate → self-check).
• Bilingual review: You see source and target side-by-side, making it easy to spot terminology inconsistencies.
• Built-in QA validator: A separate tool checks your translation for terminology errors, number consistency, and name validation—even after translation is done.
• Flexible pricing: Free tier to try; paid top-ups for higher volume. No subscription lock-in—you pay for what you use.

ChatsControl suits freelancers who want fast MTPE without buying a CAT-tool license, and small agencies that need speed and quality in one place.

Conclusion¶

Consistent terminology in MTPE isn’t optional—it’s the foundation of quality. A system combining glossary + integrated CAT tool or platform + regular updates reduces errors by 80-90% and saves 10-15% of edit time. The setup pays for itself in 2-3 projects.

Start small: 100-150 critical terms, integrate into memoQ or Trados, enable QA. Scale from there.