What Is LQA?
LQA (Linguistic Quality Assurance) is a human review of an already-finished translation: an editor reads the target text against the source and scores it against a fixed set of criteria — accuracy, fluency, terminology, and style. Unlike automated QE, LQA is done by a person, not a model, and its output is more than a score — it's a concrete list of fixes.
LQA sits at the end of the pipeline: the text gets translated first (by a person or a machine), then passes through LQA as the last filter before delivery.
Why LQA Needs to Be a Separate Step
The translator who worked on a text is rarely its best reviewer — they don't see their own recurring mistakes and are blind to context they invented themselves to resolve an ambiguous line. LQA is assigned to a second person for exactly this reason: it's a structural defense against one person's blind spots, not a bureaucratic formality.
What an LQA Editor Checks
- Accuracy — nothing lost, added, or distorted in meaning
- Fluency — the target text reads as an original, not as a translation
- Terminology — terms match the project glossary consistently throughout
- Style — tone and register match the brand guide and document type
- Local conventions — dates, currencies, units, and addresses follow the target country's rules
Each finding is typically classified the same way MQM does it: by error type and severity.
LQA and MQM: Process and Metric
LQA and MQM solve different problems and work well together:
| LQA | MQM | |
|---|---|---|
| What it is | A human review process | A system for classifying found errors |
| Output | A list of fixes and an overall verdict | A numeric score based on error weights |
| Who does it | A linguist-editor | The same editor, using a formalized scale |
In practice, an LQA editor finds the errors, and MQM (or its updated form, MQM Core) weighs them and turns the review into a number that can be compared across projects.
LQA vs. Automated Evaluation
The key difference between LQA and MTQE or other forms of QE is the human in the loop. Automated evaluation is fast and cheap but misses semantic nuance and cultural context that only a native speaker catches. Mature processes use both: QE filters out obvious failures and sets priorities, while LQA reviews whatever passed the filter or needs special attention — legal, medical, and marketing content almost always goes through LQA regardless of what the automated score says.
How to Set Up an LQA Process
- Fix the criteria upfront — the editor needs the brief, glossary, and style guide before review, not after
- Separate the roles — the translator and the LQA reviewer should not be the same person
- Use sampling wisely — full review isn't always needed; a representative sample of segments keeps large volumes under control without losing oversight
- Feed errors back into the process — a translator's systematic mistakes or recurring glossary gaps are worth fixing at the source, not just in one document
FAQ
How is LQA different from regular editing?
Regular editing improves a text stylistically. LQA checks compliance against fixed criteria and produces a formal verdict: did the text clear the quality bar or not.
Do you still need LQA if an LLM did the translation?
Yes, arguably more so: a model can produce fluent, confident-sounding text with a factual error that no automated method will catch — only a careful human who knows both languages and the context will.
Who typically performs LQA?
A second translator or a native-speaker editor who wasn't part of the original translation. For regulated industries, a subject-matter expert (a lawyer, a doctor) as well.
Can part of LQA be automated?
The mechanical part — checking terminology against the glossary, verifying numbers and dates, spelling — yes. Semantic and stylistic judgment still needs a person.
How does LQA relate to ISO standards?
Process requirements touching LQA appear in ISO 18587 (for post-editing) and ISO 5060 (for quality evaluation generally) — both give a framework, not a substitute for a company's own checklist.