Translation QA: How Professional Translators Ensure Quality
Project managers ask me some version of "how do I know the translation is good" on almost every new engagement, and it is a fair question: the person evaluating the output usually cannot read the target language well enough to check it themselves. Translation QA is the answer, but the term gets used loosely. This guide sets out what translation quality assurance actually covers, how it differs from quality control, how quality gets measured, and the specific errors a QA process has to catch on Indonesian–English work.
Mulyadi SubaliIndonesian ↔ English translator · 20+ years
What Is Translation QA?
Translation QA (TQA) is the set of processes built into a translation project to prevent errors before they reach the client. It is often confused with quality control, which is the inspection step that happens after the translation is finished.
Translation Quality Assurance (QA)
Preventive. Style guides, glossaries, translation memory checks, automated linguistic checks, and a defined workflow that catches problems while the work is being produced, not after.
Translation Quality Control (QC)
Inspective. A second linguist reviews the finished target text against the source for errors, omissions, and inconsistencies before delivery. QC catches what the preventive measures missed.
A mature workflow has both. QA reduces how much QC has to catch, and QC is the safety net for what QA cannot fully prevent. When a vendor says they have a QA process, it is worth asking which of the two they mean. My Indonesian translator vendor profile sets out how both fit into my agency workflow.
Why Translation Quality Assurance Matters
Translation errors are cheap to prevent and expensive to find late, because the client, the end reader, or a government office usually discovers them after delivery.
Rejected submissions. A mismatched name, passport number, or date in a sworn or certified translation gets the document returned by an immigration office, a court, or a university, and the deadline it was meant to meet is lost.
Inconsistency across a product line. The same term translated three ways across manuals, UI strings, and marketing copy confuses the target audience and creates support costs that have nothing to do with the original content.
Broken layouts. Translated text that runs longer than the source can overflow buttons, tables, and page layouts, which only surfaces once the file is back in production.
Lost trust. A client who cannot read the target language judges quality by the errors that reach them. One visible error makes every other deliverable suspect.
What Are the Steps in the Translation QA Process?
A sound translation quality assurance process runs in six steps: terminology alignment, translation memory checks, automated checks, self-review, independent review, and a final delivery check.
1
Style guide and glossary alignment
Before translation starts, terminology and tone are locked against the client's style guide and glossary, or new ones are built if none exist. This is the highest-leverage QA step: most downstream inconsistency traces back to terminology decisions that were unclear or never made.
2
Translation memory checks
Segments matching or resembling previously translated content are flagged automatically, so approved terminology and phrasing carry forward, and any conflict with prior work surfaces during translation rather than after delivery.
3
Automated QA checks
The CAT tool or a dedicated QA checker scans for mechanical errors that do not need human judgment: missing or mismatched numbers, terminology inconsistent with the glossary, untranslated segments, tag and formatting mismatches, double spaces. These catch what fatigue would otherwise miss.
4
Self-review against the source
The translator reviews the output segment by segment against the source, not just reading the target for fluency. Reading for fluency alone is the review method most likely to miss a meaning error that happens to read smoothly.
5
Independent review
A second linguist who has not seen the draft compares the translated text against the source material. Independent review catches what the original translator, having read their own work repeatedly, is least likely to notice.
6
Final formatting and delivery check
A last pass confirms the delivered file matches the required format, nothing broke in file conversion, and any client-specific delivery requirements were met.
How Is Translation Quality Measured? LQA and MQM
Translation quality is measured through linguistic quality assurance (LQA): a reviewer logs each error by category and severity, and the weighted total is compared against a threshold agreed with the client. The most widely used error framework is MQM (Multidimensional Quality Metrics).
Accuracy
Does the translation say what the source says? Mistranslations, omissions, additions, and untranslated text.
Fluency
Does it read correctly in the target language? Grammar, spelling, punctuation, and awkward or unidiomatic phrasing.
Terminology
Are glossary terms and domain terminology used correctly and consistently across translations?
Style
Does it follow the client's style guide, register, and brand voice, including the message and tone?
Locale conventions
Are numbers, dates, currencies, addresses, and units formatted for the target locale?
Severity
Each error is rated minor (noticeable but does not change meaning), major (changes meaning or would confuse the reader), or critical (could cause legal, safety, financial, or reputational harm). Penalty weights rise with severity.
Scored LQA is most useful on large or ongoing accounts, where quality needs to be tracked and reported across many deliveries rather than confirmed one project at a time. For smaller or one-off projects, the six-step process above without formal scoring is usually proportionate. The same scoring is also how machine translation output gets evaluated before deciding whether a content type is suitable for post-editing.
What Does QA Catch in Indonesian–English Translation?
On Indonesian–English translation and localization projects, the errors that matter most are locale formats, tense, and a handful of words whose meaning is easy to flatten. Several of them pass automated checks because the result is grammatically correct.
Number and currency format
Indonesian uses a period as the thousands separator and a comma for decimals: Rp1.500.000,50 is one and a half million rupiah and fifty sen. Copying the separators unchanged into English, or the reverse, can change an amount by a factor of a thousand. Automated number checks only catch this if the target locale is configured correctly.
Dates
Indonesian writes dates day first. 03/04/2026 is 3 April, which a US reader takes as March 4. Writing the month as a word in the English translation removes the ambiguity.
Tense
Indonesian verbs do not mark tense. Time comes from context or from markers such as sudah (already), telah, and akan (will). Translating into English forces a choice, and a wrong one can turn a completed obligation into a pending one. No automated checker flags this.
Belum vs tidak
Belum means "not yet" and implies the thing is expected; tidak is a plain "not." Rendering belum dibayar as "not paid" instead of "not yet paid" drops a meaning that can matter in a contract or a payment dispute.
Singular and plural
An Indonesian noun such as dokumen can be singular or plural. The English translation must commit to one, and in a contract the difference between "the document" and "the documents" can change scope.
Forms of address (EN→ID)
English "you" becomes Anda, Bapak/Ibu, or kamu depending on register. Mixing them across a website or app reads as careless to an Indonesian user, and it is exactly the kind of decision a style guide should settle before translation starts.
Text expansion (EN→ID)
Indonesian text often runs longer than the English source, so buttons, menus, and table headers can overflow. This is caught in an in-context review of the localized interface, not in the bilingual file.
Which Tools Are Used for Translation QA?
Translation QA relies on four kinds of tools: CAT tool QA modules, dedicated QA checkers, translation memory, and the client's glossary and style guide.
CAT tool and TMS QA modules. CAT tools such as Trados Studio and memoQ, and translation management systems such as Phrase and XTM, include automated QA checks for terminology, numbers, tags, and untranslated segments as a standard feature.
Dedicated QA checkers. Tools such as Xbench, Verifika, and lexiQA run more exhaustive checks than a built-in module, and can check consistency across a whole project or several files at once.
Translation memory. Beyond cost and speed, TM enforces consistency by showing how a term or phrase was handled in previously translated content.
Style guide and glossary. Not software, but the most consequential QA tool. A maintained glossary prevents more inconsistency than any checker catches after the fact.
Automated QA catches mechanical errors well and meaning errors poorly. A missing number or an untranslated segment gets flagged reliably. A grammatically correct sentence that says something different from the source, such as a dropped "yet" or a wrongly inferred tense, does not. That is why human review against the source is not optional for anything that matters.
What Are the Common Challenges in Translation QA?
The most common challenges are noisy automated checks, missing reference material, compressed review time, reviewers who confuse preference with error, and corrections that never feed back into the workflow.
False positives. Checkers not configured for Indonesian flag correct number formats and hyphenated plurals such as dokumen-dokumen, and a report with hundreds of false alarms gets skimmed rather than read.
No glossary or style guide. Without reference material, the reviewer and translator disagree over choices that were never defined, and the review becomes a debate rather than a check.
Review squeezed by deadlines. When a project runs late, the review step is usually what gets shortened, which is the opposite of what the risk calls for.
Preferential changes. A reviewer rewriting correct sentences to suit their own style inflates the error count and erodes trust in the process. Good LQA practice separates errors from preferences.
No feedback loop. Reviewer corrections that never make it back into the glossary and translation memory get repeated on the next project. Updating both at delivery, not just the file, is what makes quality improve over time.
What to Ask a Translator About Their QA Process
Before a large project, ask four questions: who reviews the work, what reference material exists, what the automated checks cover, and whether quality is scored.
Is there an independent reviewer, or self-review only?
Self-review alone is weaker for anything above routine, low-stakes content. Ask directly rather than assuming a second linguist is involved.
Is there a glossary and style guide for this account?
If not, ask whether one will be built before large-volume work starts. Retrofitting consistency costs more than establishing it up front.
What automated checks are run, and on what?
Confirm whether checks cover the full document or a sample, and whether the checker is configured for Indonesian number and date formats.
Is LQA scoring used, and is it available on request?
For large or ongoing accounts, a scored quality report tracks a vendor's consistency over time, not just a single delivery.
How I Handle QA on Indonesian Projects
Every project I deliver goes through glossary alignment, translation memory, automated checks, and a segment-by-segment self-review against the source, and it ships with a short QA note.
Client glossaries and termbases are loaded before the first segment, and I maintain client-specific translation memories rather than treating each project as a blank slate. I described why that matters for consistency in the piece on what translation memory actually does. The discipline comes from my years as Language Lead on the Toyota account, where maintaining the in-house terminology base was what kept technical documentation consistent from one project to the next.
Each delivery includes a QA note flagging source inconsistencies, ambiguous segments, and likely text expansion issues, so the agency's reviewer or developer knows where to look first. For certified translations, names, numbers, and dates are checked against the source document line by line, since my signed statement certifies the translation as faithful and a single mismatched digit gets the document rejected. For MTPE projects, the emphasis shifts further toward comparing every segment against the source, which I cover in the MTPE guide. Where an agency has its own reviewers, I align on the style guide and glossary with them directly rather than working from an outdated version.
A Documented QA Process for Your Indonesian Projects
Glossary from segment one, TM-based consistency checks, self-review against source, and a QA note with every delivery.
Send documents for a quote, and I'll confirm which QA steps apply to your project, and whether an independent reviewer or LQA scoring makes sense for its volume and stakes.
Translation QA refers to the processes built into a project to prevent errors before they happen: style guides, glossaries, translation memory checks, and defined workflow steps. Quality control is the inspection step after translation is complete, where a second linguist reviews the finished text against the source to catch whatever the preventive measures missed. A mature process includes both rather than treating either as sufficient on its own.
LQA, or linguistic quality assurance, is a formal and usually scored review of a translation. Reviewers log each error by category (accuracy, fluency, terminology, style, locale conventions) and by severity, commonly using a framework such as MQM (Multidimensional Quality Metrics), and the weighted result is compared against a pass threshold agreed with the client. It is most common on large or ongoing accounts where quality needs to be tracked and reported over time.
No. Automated QA tools are reliable for mechanical errors such as missing numbers, untranslated segments, inconsistent terminology, and tag or formatting mismatches. They cannot reliably catch meaning-level errors, such as a dropped negation or a wrongly inferred tense, when the resulting sentence is grammatically correct and reads smoothly. That is why an independent human review against the source remains necessary for anything beyond routine, low-stakes content.
The most consequential ones are number and date formats (Indonesian uses a period as the thousands separator and a comma for decimals, and writes dates day-first), tense, which Indonesian verbs do not mark and the translator must infer from context, and the difference between "belum" (not yet) and "tidak" (not). In the English-to-Indonesian direction, inconsistent forms of address and text expansion in user interfaces are the recurring issues.
Yes, particularly for large or ongoing projects. Ask whether there is an independent reviewer or only self-review, whether a glossary and style guide exist for the account, what automated checks are run and whether they cover the full document, and whether LQA scoring is available if you need to track quality over time rather than just confirm a single delivery.
Some level of QA belongs on every project: self-review against the source and basic automated checks take little additional time. Independent review and formal LQA scoring are where scale matters. They can be disproportionate for a routine, low-stakes document, but are worth insisting on for anything high-volume, high-visibility, or carrying real consequence if an error slips through.
Mulyadi Subali
Indonesian ↔ English translator and interpreter. Former Language Lead at the World Bank, UOB, Ford and Toyota; legal translation certification from Universitas Indonesia in both directions; ProZ Certified PRO since 2009.