Linguistic quality assurance (LQA) is the process of reviewing translated content against a defined set of error categories and severity levels, so translation quality can be measured, scored and improved instead of judged by opinion. A reviewer marks each error, for example accuracy, terminology or style, assigns a severity, and the result is a quality score that decides whether a translation passes. This guide explains how LQA works, the most common frameworks, how to calculate a score and how to build an LQA process that scales.
LQA vs QA vs Review
| Activity | What it checks | Output |
|---|---|---|
| Automated QA | Numbers, tags, placeholders, terminology, spelling, consistency | List of potential issues to fix |
| Review / revision | A second linguist edits the translation | An improved translation |
| Linguistic quality assurance | A sample or full text is evaluated against an error typology | A quality score, error statistics and feedback |
Automated checks and LQA work best together. Automated QA catches mechanical errors on every segment, while LQA measures overall quality and shows where translators and engines need to improve. Our translation quality assurance checklist covers the automated side.
LQA Frameworks: MQM and ISO 5060
The most widely used framework is MQM (Multidimensional Quality Metrics), an open error typology used by many LSPs, enterprises and machine translation researchers. In 2024, ISO published ISO 5060, which gives general guidance for evaluating translation output with an analytic, error-based approach. Most company LQA models are based on these standards with their own adjustments.
Typical LQA Error Categories
- Accuracy: mistranslation, omission, addition, untranslated text.
- Terminology: wrong or inconsistent use of approved terms.
- Linguistic conventions: grammar, spelling, punctuation.
- Style: tone, register and compliance with the style guide.
- Locale conventions: dates, numbers, currencies, address formats.
- Design and markup: truncation, broken tags, layout issues.
Severity Levels
| Severity | Meaning | Typical penalty points |
|---|---|---|
| Minor | Noticeable, but meaning and usability are not affected | 1 |
| Major | Meaning is affected or the text is confusing | 5 |
| Critical | Could cause legal, safety, financial or reputational harm | 10 or more, often an automatic fail |
Penalty values vary by organization. What matters is that everyone uses the same definitions.
How to Calculate an LQA Score
- Choose a sample, for example 1,000 to 2,000 words, or review the full file for high-risk content.
- Mark each error with a category and severity.
- Add up the penalty points.
- Normalize by word count, for example penalty points per 1,000 words.
- Compare the result with your pass threshold, which can differ by content type.
For example, 2 minor errors and 1 major error in a 1,000-word sample give 7 penalty points. If your threshold is 10 points per 1,000 words, the translation passes.
How to Build an LQA Process
- Define quality per content type. Legal texts and UI strings need stricter thresholds than internal help articles.
- Agree on a shared error typology and train reviewers with examples.
- Provide reference material: glossary, style guide, translation memory.
- Score regularly, not only when there are complaints.
- Close the loop. Share feedback with translators and update terminology and translation memory.
- Track trends per language, vendor and AI engine.
LQA for Machine Translation and AI
LQA is the most reliable way to compare AI engines on your own content. Score samples from DeepL, GPT, Claude or other engines with the same typology, and you will see which engine performs best for each language and domain. The same scores help you decide how much machine translation post-editing each content type needs.
LQA in WordBeam
WordBeam combines automated QA checks and AI proofreading with LQA scorecards on the Studio plan, so reviewers can score translations by error category and severity alongside the work in the CAT editor. Approved corrections are saved to translation memory and the term base, so the same errors are not repeated. See the plans on WordBeam pricing and start a free 14-day trial.
Frequently Asked Questions
What does LQA stand for in translation?
LQA stands for linguistic quality assurance, a structured evaluation of translation quality using error categories and severity levels.
What is MQM?
MQM, or Multidimensional Quality Metrics, is an open framework that defines error types and severities for evaluating translations.
How often should you run LQA?
Regularly. Many teams score a sample of every project for high-risk content and a periodic sample for lower-risk content.
Can AI perform LQA?
AI can pre-flag likely errors and speed up review, but final scoring for important content is still best done by qualified linguists.