What is LQA? Complete Guide to Linguistic Quality Assurance
Linguistic Quality Assurance (LQA) is a systematic process for evaluating the quality of translated content. It goes beyond simple proofreading to provide objective, measurable assessments of translation accuracy, fluency, and overall fitness for purpose.
This guide covers everything about LQA: what it is, how it works, and how to implement it in your translation workflow.
What is LQA?
LQA (Linguistic Quality Assurance) is the process of evaluating translated content against defined quality standards. Unlike general proofreading, LQA uses structured methodologies — typically based on error typologies like MQM — to identify, categorize, and score translation errors.
LQA vs. Proofreading vs. Editing
| Activity | Focus | Output |
|---|---|---|
| Proofreading | Fixing surface errors | Corrected text |
| Editing | Improving style and clarity | Better text |
| LQA | Evaluating quality objectively | Quality score + error report |
The key difference: LQA measures quality rather than just fixes problems. That measurement is what makes everything else possible — comparing vendor performance, tracking quality trends over time, giving translators concrete feedback, and verifying SLA compliance.
Why LQA Matters
Without LQA, you're guessing about quality. Here's what's at stake.
Quality Consistency
Left unchecked, quality varies wildly between translators, projects, and languages. One translator might be excellent, another average, and you'd have no objective way to tell. LQA creates a shared standard everyone works against.
Cost Control
Bad translations are expensive — but not in the ways you might expect. Sure, there's rework. But there's also customer complaints, returns, brand erosion, and in regulated industries, potential legal exposure. A single mistranslated medical instruction can cost more than a year's worth of LQA. Catching issues early is just cheaper.
Vendor Management
When you're working with five different agencies across twelve languages, opinions about "good enough" vary. LQA gives you numbers instead of arguments.
Compliance
Healthcare, legal, and finance all require documented quality processes. LQA provides the audit trail regulators want to see.
Continuous Improvement
By tracking error patterns — say, a translator who consistently struggles with passive constructions in German — LQA identifies specific issues you can fix through training or process changes.
The LQA Process
A typical LQA workflow has six steps. Not every organization runs all of them, but this is the full picture.
Step 1: Define Quality Criteria
Before anyone evaluates anything, establish what you're measuring:
- Error categories to track (accuracy, fluency, terminology, etc.)
- Severity levels (critical, major, minor)
- Passing threshold (e.g., MQM score of 95 or above)
- Sample size (100% review or a representative sample)
Step 2: Select Evaluators
LQA evaluators should be native speakers of the target language, ideally with subject matter expertise. They need training in the LQA methodology, and — this part matters — they should be independent from the original translators. You don't want people grading their own work.
Step 3: Perform Evaluation
Evaluators review translations segment by segment. For each error they find, they record the error type (mistranslation, omission, grammar, etc.), severity (how much impact does it have?), and location (which segment?).
This is the most time-consuming step by far.
Step 4: Calculate Quality Score
Using a scoring model like MQM:
Quality Score = 100 - (Penalty Points / Word Count × 100) Where penalty points depend on error severity:
- Critical: 25 points
- Major: 5 points
- Minor: 1 point
Step 5: Generate Reports
LQA reports typically include the overall quality score, error breakdown by category and severity, specific error annotations with examples, and comparison with historical performance.
A good report doesn't just say "this scored 94." It tells you why.
Step 6: Feedback Loop
Share LQA findings with translators. Focus on patterns in error types, give specific examples with correct alternatives, and — people forget this — recognize high-quality work too. LQA shouldn't feel like punishment.
LQA Error Categories
Based on the MQM framework, common LQA error categories include:
Accuracy Errors
| Error Type | Description | Example |
|---|---|---|
| Mistranslation | Meaning incorrectly conveyed | "annual" → "monthly" |
| Omission | Content missing from translation | Skipped sentence |
| Addition | Extra content not in source | Unexplained additions |
| Untranslated | Source text left as-is | English term in Spanish text |
Fluency Errors
| Error Type | Description | Example |
|---|---|---|
| Grammar | Grammatical mistakes | "The datas is..." |
| Spelling | Misspelled words | "recieve" |
| Punctuation | Incorrect punctuation | Missing comma |
| Typography | Font, spacing issues | Double spaces |
Terminology Errors
| Error Type | Description | Example |
|---|---|---|
| Wrong term | Incorrect terminology | "mouse" as animal vs. device |
| Inconsistency | Same term translated differently | Varying product names |
| Unapproved term | Term not in client glossary | Using alternative without approval |
Style Errors
| Error Type | Description | Example |
|---|---|---|
| Register | Wrong formality level | "you" vs. formal equivalent |
| Unidiomatic | Awkward phrasing | Literal translation that sounds wrong |
| Inconsistent style | Varying tone within document | Mixing formal and casual |
Locale Errors
| Error Type | Description | Example |
|---|---|---|
| Date format | Wrong date convention | 12/31/2025 vs. 31/12/2025 |
| Number format | Wrong decimal/thousand separator | 1.000 vs. 1,000 |
| Currency | Incorrect currency handling | Wrong symbol or format |
LQA Severity Levels
Critical Errors
These are the ones that keep localization managers up at night. They cause legal liability, safety risks, financial loss, or severe misunderstanding.
Examples: Medical dosage errors, legal term mistakes, safety instruction omissions.
Typical Penalty: 25 points
Major Errors
Errors that noticeably hurt comprehension, user experience, or professional appearance. A reader would stop and think "that's wrong."
Examples: Wrong meaning, confusing sentence structure, inappropriate tone.
Typical Penalty: 5 points
Minor Errors
Noticeable but don't really affect understanding. The kind of thing a careful reader catches but that doesn't change the message.
Examples: Minor punctuation, slight awkwardness, capitalization.
Typical Penalty: 1 point
LQA Metrics and KPIs
Quality Score (MQM-based)
The primary metric:
Score = 100 - (Total Penalty / Word Count × 100) Error Rate
Errors per 1000 words:
Error Rate = (Total Errors / Word Count) × 1000 Pass Rate
Percentage of translations meeting the quality threshold:
Pass Rate = (Passing Translations / Total Translations) × 100 Error Distribution
Breakdown of errors by category (accuracy, fluency, etc.), severity (critical, major, minor), and translator or vendor. This is where the real diagnostic value lives.
AI-Powered LQA
AI is changing how LQA gets done. Not replacing it — changing it.
Traditional LQA vs. AI LQA
| Aspect | Traditional LQA | AI LQA |
|---|---|---|
| Speed | Hours per document | Minutes per document |
| Cost | High (human evaluator time) | Lower per evaluation |
| Consistency | Varies by evaluator | Highly consistent |
| Scalability | Limited | Virtually unlimited |
| Subtlety | Excellent | Good and improving |
How AI LQA Works
Modern AI LQA tools use large language models to compare source and target texts, identify potential errors, classify them by type and severity, calculate quality scores, and generate detailed reports.
The speed difference is dramatic. What takes a human evaluator half a day, AI can do in minutes. But speed isn't everything.
AI LQA Limitations
AI LQA is powerful, but it's not perfect:
- It can miss subtle cultural references
- It struggles with highly creative content (marketing taglines, literary translation)
- Critical content still needs human eyes
- Regulated industries can't rely on AI alone for final quality decisions
Honestly, anyone telling you AI can fully replace human LQA evaluators today is overselling. It's a tool, not a replacement.
Best Practice: Hybrid Approach
The smart play is combining both:
- AI first — Quick initial assessment at scale
- Human review — Verify AI findings, especially critical errors
- Random sampling — Human spot-checks on AI-passed content
- Continuous calibration — Use human feedback to improve AI accuracy
Implementing LQA in Your Organization
Step 1: Choose Your Framework
Pick an error typology:
- MQM — Industry standard, highly customizable
- LISA QA — Legacy but still in use
- Custom — Based on your specific needs
Most organizations should start with MQM. Why reinvent the wheel?
Step 2: Define Quality Tiers
Not all content needs the same level of scrutiny:
| Tier | Content Type | LQA Intensity | Pass Threshold |
|---|---|---|---|
| Premium | Legal, medical, marketing | 100% review | 98+ |
| Standard | Business, documentation | 20% sample | 95+ |
| Basic | Internal, user-generated | AI-only | 90+ |
Step 3: Select Tools
LQA tools range from spreadsheets (simple but manual) to dedicated LQA platforms (purpose-built for translation QA) to TMS-integrated solutions to AI-powered platforms like KTTC with automated LQA.
Step 4: Train Your Team
Make sure evaluators understand error category definitions, severity criteria, tool usage, and calibration processes. Don't skip this. Untrained evaluators produce unreliable scores, and unreliable scores are worse than no scores at all.
Step 5: Establish Calibration
Regular calibration keeps evaluators aligned. Have them evaluate the same content independently, compare results, discuss discrepancies, and update guidelines based on what you learn.
FAQ
What does LQA mean in translation?
LQA stands for Linguistic Quality Assurance. It's the process of systematically evaluating translation quality using standardized error categories, severity levels, and scoring systems. LQA provides objective, measurable quality assessments rather than subjective opinions.
What is the difference between LQA and QA?
QA (Quality Assurance) is a broad term covering all quality-related activities. LQA specifically focuses on linguistic aspects of translation quality — accuracy, fluency, terminology, style, and locale conventions. Technical QA might cover formatting, functionality, or user experience issues that aren't linguistic.
How is LQA score calculated?
LQA scores are typically calculated using the MQM (Multidimensional Quality Metrics) model. Errors are identified, categorized by type and severity, and assigned penalty points. The score equals 100 minus the total penalty divided by word count times 100. For example: 100 - (15 penalty points / 1000 words × 100) = 98.5.
What is a good LQA score?
It depends on the content type. Generally: 99-100 is excellent (publishable as-is), 95-98 is good (minor review needed), 90-94 is acceptable (corrections required), and below 90 typically requires significant revision. Critical content like legal or medical documents often requires 98+.
Can AI replace human LQA evaluators?
Not entirely. AI LQA is excellent for initial screening, consistency, and scale, but human evaluators remain essential for subtle judgment, cultural adaptation assessment, and validation of critical content. The best approach right now is a hybrid — AI for speed and coverage, humans for depth and judgment.
What's Next
LQA isn't optional for professional translation. Whether you're using human evaluators, AI tools, or both, the key is having a systematic, measurable process that produces actionable results — not just scores, but insights you can act on.
If you don't have an LQA process today, start small. Pick a framework, define your severity levels, and evaluate a few projects. You'll learn more from those first evaluations than from months of planning.
Ready to implement LQA in your translation workflow? Try KTTC for AI-powered linguistic quality assurance with MQM-based evaluation and detailed error reporting.