What is a Fuzzy Match?
A fuzzy match occurs when a translation memory (TM) contains a segment that is similar but not identical to the source text being translated. Unlike a 100% match (exact match), fuzzy matches require the translator to review and adapt the suggested translation to fit the new context.
Fuzzy match percentages typically range from 50% to 99%, indicating how similar the stored segment is to the current source text.
Match Percentage Levels
Match Types
| Match Level | Description | Translator Action |
|---|---|---|
| 100% (Exact) | Identical source text | Verify, usually accept |
| 101% (Context) | 100% + same surrounding text | Highest confidence |
| 95-99% | Very similar, minor differences | Quick edit |
| 85-94% | Similar, some changes needed | Moderate edit |
| 75-84% | Partially similar | Significant edit |
| 50-74% | Low similarity | Heavy edit or retranslate |
| <50% | Minimal similarity | Usually not shown |
Example of Fuzzy Matches
Original TM entry (100%):
"Click the Save button to save your changes."
New source text vs. TM match:
| New Text | Match % | Difference |
|---|---|---|
| "Click the Save button to save your changes." | 100% | None |
| "Click the Submit button to save your changes." | 92% | One word |
| "Click the Save button to apply your changes." | 88% | One word |
| "Click the OK button to confirm your changes." | 78% | Two words |
| "Press Save to store changes." | 65% | Restructured |
How Fuzzy Matching Works
Algorithm Basics
CAT tools use various algorithms to calculate similarity:
Levenshtein Distance
- Counts minimum edits (insertions, deletions, substitutions)
- Character-level or word-level comparison
Token-Based Matching
- Compares individual words/tokens
- Considers word order
N-gram Analysis
- Compares sequences of characters/words
- Better for structural changes
Calculation Example
Source: "Save your document before closing"
TM: "Save your file before closing"
Tokens match: 4/5 = 80%
With word importance weighting: ~85%
Fuzzy Match Pricing
Industry Standard Rates
| Match Range | Typical Discount |
|---|---|
| 100% | 70-90% off |
| 95-99% | 50-75% off |
| 85-94% | 30-50% off |
| 75-84% | 15-30% off |
| 50-74% | 0-15% off |
| No match | Full rate |
Pricing Considerations
- Higher matches = less editing effort
- Quality of TM affects actual effort
- Domain complexity matters
- Some clients negotiate different bands
Working with Fuzzy Matches
Best Practices for Translators
Always Review
- Never accept fuzzy matches blindly
- Check for subtle meaning changes
- Verify terminology consistency
Efficient Editing
- Focus on highlighted differences
- Use CAT tool comparison views
- Leverage concordance search
Common Pitfalls
- False confidence in high percentages
- Missing context changes
- Overlooking number/date differences
Example Workflow
Source: "Enter your email address to continue."
TM Match (87%): "Enter your email address to proceed."
^^^^^^^^
Suggestion: "Введите адрес электронной почты, чтобы продолжить."
^^^^^^^^^
Action: Change "proceed" translation to "continue" equivalent
Fuzzy Match Penalties
CAT tools can apply penalties to reduce match scores:
Common Penalty Types
| Penalty Type | Purpose | Typical Reduction |
|---|---|---|
| Different TM | Lower trusted source | -5 to -10% |
| Old translation | Outdated content | -5% |
| Different project | Context mismatch | -3 to -5% |
| Different file type | Format concerns | -2 to -5% |
| Machine origin | MT-generated | -10 to -20% |
Why Use Penalties?
- Prioritize trusted sources
- Account for context differences
- Manage multiple TM quality levels
- Distinguish MT from human translation
Fuzzy Match Thresholds
Configuring Minimum Thresholds
Most CAT tools allow setting minimum fuzzy match thresholds:
| Setting | Effect |
|---|---|
| 70% minimum | Shows more suggestions, more noise |
| 75% minimum | Balanced, common default |
| 80% minimum | Fewer but higher quality matches |
| 85% minimum | Conservative, only useful matches |
Threshold Strategy
- High-volume projects: Lower threshold (more matches)
- Quality-critical: Higher threshold (less noise)
- New TM: Lower threshold (build matches)
- Mature TM: Higher threshold (trust quality)
Impact on Productivity
Productivity Gains by Match Level
| Match Level | Words/Hour Increase |
|---|---|
| 100% | +300-400% |
| 95-99% | +200-300% |
| 85-94% | +100-150% |
| 75-84% | +50-75% |
| No match | Baseline |
ROI Calculation
For 100,000 words with typical TM leverage:
| Match Level | Words | Time Saved |
|---|---|---|
| 100% | 20,000 | 80% |
| 95-99% | 15,000 | 60% |
| 85-94% | 10,000 | 40% |
| 75-84% | 5,000 | 20% |
| <75% | 50,000 | 0% |
Overall project savings: ~35%
FAQ
What is a good fuzzy match percentage?
Matches above 75% are generally useful. Matches 85%+ provide significant productivity benefits. Below 75%, the editing effort may equal or exceed translation from scratch.
Should I accept high fuzzy matches without review?
No. Even 99% matches can have critical differences — a single changed word might completely alter the meaning. Always review fuzzy matches, regardless of percentage.
How do CAT tools calculate fuzzy match percentages?
Different tools use different algorithms (Levenshtein distance, token matching, n-grams). The same segment pair may show different percentages in different tools. Focus on the actual differences, not just the number.
Can fuzzy matches introduce errors?
Yes. Common issues include:
- Accepting inappropriate matches
- Missing subtle meaning changes
- Propagating errors from original translation
- Inconsistent terminology
How do fuzzy matches affect translation memory building?
When you edit a fuzzy match and confirm it, the new segment is typically added to the TM. Over time, this builds up exact matches for content that was previously only fuzzy matched.
What's the difference between fuzzy match and machine translation?
Fuzzy matches come from human-translated content in your TM. Machine translation is generated by AI/algorithms. Fuzzy matches are typically more reliable but limited to similar previously translated content.