A translation can be accurate and still fail. A button label that is too long for the screen, a price written with the wrong decimal separator or a product name translated three different ways all reach users as bugs. Translation quality assurance is the process that catches these problems before release, and it works best when it is a routine checklist rather than a last-minute scramble.
Below is a practical checklist you can reuse for every release, whether you ship an app every week or publish a new website version once a quarter.
What Is Translation Quality Assurance?
Translation quality assurance (QA) is a set of checks that confirm translated content is correct, consistent and ready to publish. It combines automated checks, which catch mechanical errors, with human review, which judges meaning, tone and cultural fit.
QA is different from quality control. Quality control looks at a finished translation and asks whether it is good. Quality assurance covers the whole process, including the glossary, the reference material and the workflow, so fewer errors are created in the first place.
Why QA Matters More in 2026
With machine translation and LLMs producing first drafts, volume has gone up and the type of errors has changed. AI output is fluent, which makes mistakes harder to spot by eye. Omitted sentences, invented details and inconsistent terminology can slip through when reviewers read quickly. A structured QA step is what keeps faster workflows safe.
The Translation QA Checklist
1. Before translation starts
- The source text is final, or changes are clearly tracked.
- A current glossary or term base is attached to the project.
- A style guide defines tone, formality and how to address the user in each language.
- Translators have context: screenshots, character limits and notes on placeholders.
- Translation memory is connected so approved wording is reused.
2. Automated checks
- All placeholders and variables (such as
{name}or%s) are present and unchanged. - HTML and XML tags are complete and in the right order.
- Numbers, dates, currencies and units match the source and follow local formats.
- Glossary terms are used and forbidden terms are absent.
- No segments are left empty or untranslated.
- Text length stays within UI limits.
- Punctuation, leading and trailing spaces and double spaces are consistent.
- Identical source segments have identical translations, unless context requires otherwise.
These checks are repetitive, so they should run automatically inside your CAT tool every time a segment is confirmed, not once at the end.
3. Linguistic review
- Meaning is complete, with nothing added or left out.
- Grammar, spelling and punctuation are correct for the target locale.
- Tone matches the style guide and the brand.
- Terminology is consistent across the whole project, not only within one file.
- Cultural references, examples and idioms work for the target market.
4. In-context review
- Translations are checked on the real screen, page or document layout.
- No text is truncated, overlapping or wrapping badly.
- Right-to-left languages display correctly.
- Links, buttons and forms work in every language version.
5. Before release
- All open reviewer comments are resolved.
- Corrections are saved back to translation memory.
- New terms are added to the glossary.
- A short QA report records error counts and categories for each language.
How to Measure Translation Quality
Saying a translation is "good" or "bad" does not help anyone improve. A shared error typology does. Many teams use a model based on MQM (Multidimensional Quality Metrics), which groups issues into categories such as accuracy, fluency, terminology, style and locale conventions, each with a severity level.
| Severity | Example | Typical action |
|---|---|---|
| Critical | Wrong price, unsafe instruction, broken placeholder | Block release |
| Major | Meaning changed, wrong product term | Fix before release |
| Minor | Awkward phrasing, small punctuation issue | Fix when possible |
| Preferential | Reviewer would choose a different synonym | Log, do not count as error |
Scoring with weights per severity turns review into numbers you can track over time and compare between vendors, engines and languages. For process requirements on the human side, such as revision by a second linguist, the ISO 17100 standard for translation services is a useful reference.
Common QA Mistakes
- Running QA only at the end. Errors are cheaper to fix while the translator is still in the file.
- Ignoring false positives. If a QA tool flags hundreds of irrelevant warnings, people stop reading them. Tune the rules.
- No feedback loop. If corrections never reach the translation memory or glossary, the same errors return in the next release.
- Reviewing without context. A reviewer who never sees the screen cannot judge truncation or meaning in the UI.
Good terminology is the foundation of most of these checks. Our guide on terminology management and translation glossaries shows how to build one that people actually use. If your team edits AI output, see also machine translation post-editing.
How WordBeam Helps
WordBeam runs automated QA checks for tags, placeholders, numbers and terminology directly in the CAT editor, together with AI proofreading that highlights meaning and fluency issues. LQA scorecards let reviewers log errors by category and severity, and every fix updates translation memory automatically, so quality improves with each release.
Frequently Asked Questions
What is the difference between translation QA and LQA?
Translation QA usually refers to the full set of checks, including automated ones. LQA (linguistic quality assurance) focuses on the human evaluation of language quality, often with an error scorecard.
Can QA be fully automated?
Mechanical checks can be. Judging meaning, tone and cultural fit still needs a qualified human reviewer, especially for customer-facing content.
How much of a project should be reviewed?
It depends on risk. High-visibility and regulated content should be reviewed in full. For lower-risk content, many teams review a sample and rely on automated checks for the rest.