“The translation scored 92%.” What does that actually mean?

Quality metrics in translation suffer from a fundamental problem: they’re often abstract numbers disconnected from what matters. A project can score well on mechanical metrics while producing translations that don’t work for their intended purpose. Or score poorly on pedantic criteria while delivering effective communication.

Better metrics connect quality assessment to actual outcomes.

The problem with single-number scores

A single quality score collapses complex information into one figure:

• 500 segments
• 12 terminology errors
• 3 grammar issues
• 2 meaning changes
• 97 style variations
• … somehow equals “87%”

What should you do with this? Is 87% good? The score doesn’t tell you what’s wrong, what’s important, or what to fix first.

Worse, it doesn’t tell you whether those issues matter for this specific content. A style variation in a legal contract is irrelevant; in brand marketing, it might be a problem. Generic scores can’t distinguish.

Structured quality breakdowns

Replace the single score with structured breakdowns:

By error category:

• Accuracy: 98.2%
• Fluency: 94.6%
• Terminology: 91.3%
• Style: 89.7%

Now you can see that accuracy is strong but style needs attention.

By severity:

• Critical errors: 0
• Major errors: 4
• Minor errors: 18

Zero critical errors means no blocking issues. The major errors need review.

By segment:

• Segments above threshold: 487 (97.4%)
• Segments needing review: 13 (2.6%)

You know exactly where to focus attention.

Content-type appropriate metrics

Different content types need different quality standards:

Legal content:

• Accuracy is critical—meaning changes are unacceptable
• Terminology must match established legal language exactly
• Style variations are acceptable if meaning is preserved
• Key metric: accuracy score, terminology compliance

Marketing content:

• Fluency and naturalness matter most
• Strict accuracy may be less important than impact
• Style should match brand guidelines
• Key metric: fluency score, brand voice alignment

Technical documentation:

• Terminology consistency is crucial
• Accuracy of technical information is essential
• Readability ensures users can follow instructions
• Key metric: terminology consistency, technical accuracy

UI strings:

• Length constraints must be met
• Consistency with existing UI translations
• Terminology alignment with product
• Key metric: length compliance, consistency score

Applying legal-content standards to marketing copy produces irrelevant metrics. The framework should match the content.

Trending and baselines

Point-in-time metrics are less useful than trends:

Baseline establishment: What’s the normal quality level for this content type, language pair, and process?

Variance tracking: When does quality deviate from baseline? Up or down?

Pattern identification: Do quality problems correlate with volume, deadline pressure, specific source content?

Improvement measurement: Are process changes producing better results?

A score of 92% is meaningless in isolation. A score of 92% when the baseline is 89% indicates improvement. A score of 92% when the baseline is 96% signals a problem.

Vendor and source comparison

When multiple sources produce translations, comparative metrics reveal performance differences:

Vendor comparison:

• Vendor A average quality: 94.2%
• Vendor B average quality: 91.8%
• Vendor C average quality: 93.5%

Error profile comparison:

• Vendor A: Strong accuracy, weak style
• Vendor B: Consistent but slow on terminology updates
• Vendor C: Highest fluency, occasional accuracy issues

Language pair performance:

• EN→DE: All vendors strong
• EN→JA: Vendor A significantly better
• EN→PT: Vendor B leads

This data enables informed sourcing decisions—not just “who’s cheapest” but “who produces best results for which content.”

Cost-quality relationships

Quality metrics should connect to economics:

Error resolution cost: What does it cost to fix each error type? (Major errors requiring SME review cost more than minor grammar fixes.)

Quality-adjusted cost: Raw per-word rates don’t account for post-editing. A cheaper vendor requiring heavy editing may cost more net.

Quality threshold economics: At what quality level is content publishable without review? What volume falls below that threshold?

These connections enable optimization: invest in quality improvement where error costs are high, accept lower quality where post-editing is cheaper than perfection.

Actionable quality data

Metrics should drive action:

Immediately actionable:

• “Segment 47 has a critical accuracy error” → Fix it now
• “Terminology XYZ used incorrectly 8 times” → Batch correction needed

Process improvement:

• “Accuracy errors increased 15% this month” → Investigate cause
• “Vendor X terminology compliance declining” → Address with vendor

Strategic decisions:

• “Japanese quality consistently 10 points below German” → Different approach needed for Japanese
• “Technical content QA taking 3x marketing content” → Adjust expectations or process

Metrics that don’t connect to action are just reporting for its own sake.

Automated metric generation

Manual metric tracking doesn’t scale. Automated systems should generate metrics as byproduct of QA:

Per-segment: Quality scores, error flags, severity levels Per-project: Aggregate scores, error distribution, segment pass rates Per-period: Trends, comparisons, anomaly detection Per-dimension: By language, vendor, content type, date range

Dashboards present this data accessibly. Alerts notify when metrics cross thresholds. Reports enable quality reviews.

The measurement investment

Good quality metrics require:

• Clear definition of what quality means for your content
• Structured error taxonomy applied consistently
• Systems that capture and aggregate data
• Baseline establishment for meaningful comparison
• Regular review and action on metric findings

This investment produces compounding returns: better vendor management, targeted process improvement, defensible quality claims, and actual quality improvement over time.

Language Ops provides structured quality metrics with category breakdowns, trend analysis, and comparative reporting. See quality analytics on your translation data.