What is a Segment?
A segment is a discrete unit of text that is handled as a single translatable entity in CAT tools and translation memory systems. Segmentation is the process of dividing source content into these manageable units, which then become the foundation for translation, storage, and reuse.
The most common segment boundary is the sentence, but segments can also be paragraphs, headings, list items, or other logical text divisions depending on the content type and segmentation rules applied.
Segmentation Types
Sentence-Based Segmentation
The most common approach, breaking text at sentence boundaries.
Example:
Source text:
"Welcome to our platform. We help you translate content efficiently."
Segments:
1. "Welcome to our platform."
2. "We help you translate content efficiently."
Advantages:
- Natural translation units
- Good TM leverage
- Industry standard
Challenges:
- Abbreviations (Dr., Inc., etc.)
- Numbers with periods (3.14)
- Ellipsis (...)
- Language-specific patterns
Paragraph-Based Segmentation
Each paragraph becomes one segment.
Best for:
- Literary translation
- Marketing content
- Content requiring flow preservation
Drawbacks:
- Lower TM reuse
- Larger segments to manage
Custom Segmentation
Based on specific rules or patterns:
- By HTML tags
- By XML elements
- By line breaks
- By custom delimiters
Segmentation Rules
Standard Breaking Points
| Pattern | Action | Example |
|---|---|---|
| Period + space | Break | "End. Start" → 2 segments |
| Question mark + space | Break | "Why? Because" → 2 segments |
| Exclamation + space | Break | "Stop! Continue" → 2 segments |
| Colon (contextual) | Sometimes | Depends on rules |
| Semicolon | Usually not | Stays in segment |
Exception Handling
Abbreviations:
"Dr. Smith arrived at 9 a.m. today."
→ One segment (abbreviations don't break)
Numbers:
"The result was 3.14159."
→ One segment (decimal doesn't break)
Quoted text:
"He said: 'Hello. How are you?'"
→ One or two segments (depends on rules)
Segmentation in CAT Tools
SRX (Segmentation Rules eXchange)
The standard format for defining segmentation rules:
<srx version="2.0">
<body>
<languagerules>
<languagerule languagerulename="English">
<rule break="no">
<beforebreak>\b(Dr|Mr|Mrs|Ms)\.</beforebreak>
<afterbreak>\s</afterbreak>
</rule>
<rule break="yes">
<beforebreak>[.?!]+</beforebreak>
<afterbreak>\s+[A-Z]</afterbreak>
</rule>
</languagerule>
</languagerules>
</body>
</srx>
Tool-Specific Rules
| CAT Tool | Segmentation Approach |
|---|---|
| Trados | Built-in + custom SRX |
| memoQ | Segmentation rules editor |
| Phrase | Project-level configuration |
| OmegaT | SRX files |
Impact on Translation Memory
Segment Granularity vs Reuse
| Granularity | TM Reuse | Context Preservation |
|---|---|---|
| Sentence | High | Low |
| Paragraph | Low | High |
| Sub-sentence | Very High | Very Low |
Example Impact
Sentence segmentation:
TM Entry: "Click Save to continue."
New text: "Click Save to continue."
Match: 100%
Paragraph segmentation:
TM Entry: "Click Save to continue. Your changes will be applied."
New text: "Click Save to continue. Your settings will be saved."
Match: ~70%
Best Practices
Consistent Segmentation
Use same rules across project
- Ensures TM consistency
- Enables accurate matching
Document your rules
- Share with team members
- Include in project setup
Consider content type
- Technical: sentence-level
- Marketing: may need paragraph
- Software: string-by-string
Handling Problematic Segments
Too long segments:
- May need manual splitting
- Review segmentation rules
- Consider content restructure
Too short segments:
- Merge if appropriate
- Adjust breaking rules
- Check for over-segmentation
Segments with inline codes:
Original: "Click <b>here</b> to start."
Segment: "Click <b>here</b> to start."
(Inline formatting preserved within segment)
Segment Status and Workflow
Common Segment States
| State | Meaning |
|---|---|
| New | Not yet translated |
| Draft | Translation in progress |
| Translated | Translation complete |
| Reviewed | Passed review |
| Confirmed | Final, locked |
| Rejected | Needs rework |
Workflow Example
New → Translated → Reviewed → Confirmed
↓
Rejected → Translated (revised)
Segment-Level Features
Segment Properties
- Source text
- Target text
- Match percentage
- Origin (TM, MT, human)
- Status
- Comments/notes
- Metadata (created, modified, by whom)
Segment Operations
| Operation | Purpose |
|---|---|
| Split | Divide into multiple segments |
| Merge | Combine multiple segments |
| Lock | Prevent changes |
| Copy source | Fill target with source |
| Insert match | Use TM suggestion |
FAQ
What is the ideal segment size?
Sentence-level is the industry standard for most content types. It balances TM leverage with translation context. However, marketing or literary content may benefit from paragraph-level for better flow.
Can I change segmentation after starting translation?
Yes, but with caution. Re-segmenting can affect TM matches and may require re-translation of affected segments. It's best to finalize segmentation rules before starting.
Why do different CAT tools segment the same text differently?
Different tools use different segmentation algorithms and default rules. Even with SRX standards, implementations vary. Always verify segmentation when switching tools.
How do segments relate to translation memory?
Each confirmed segment becomes a TM entry (translation unit). When new text is segmented the same way, TM can find matches. Consistent segmentation is crucial for TM effectiveness.
What happens to inline formatting in segments?
Inline tags (bold, italic, links) are preserved within segments as placeholders or codes. Translators must maintain them in the target while translating the text around them.
Should I segment by sentence or paragraph for legal documents?
Generally sentence-level for contracts and legal text, as it provides better TM leverage and precise matching. However, some legal clauses may need to stay together for context.