Segmentation tool

The Segmentation tool marks sentence or chunk boundaries within each block. It records them as a stand-off overlay and leaves the block's runs unchanged, so dropping the overlay restores the unsegmented block. --engine selects the segmenter:

srx, the default, finds sentence boundaries with built-in rules and needs no configuration. rulesPath loads an SRX 2.0 rules file in their place.
uax29 applies the Unicode default sentence boundaries, with no language-specific exceptions.
llm asks an AI provider for meaning-based chunks rather than sentences.

Plugins can register further engines, which appear as options. Bilingual formats write the sentence layer as segments. A block that already has a layer keeps it unless overwriteSegmentation is set.

IDsegmentation

SourceBuilt-in

CategoryText processing

Cardinalitymonolingual

Tagstext-processing

Parameters

Parameter	Type	Default	Description
`engine`	string	srx	Which segmenter to use; the default needs no configuration
`layer`	string		Segmentation overlay layer name; empty uses the engine's natural layer
`overwriteSegmentation`	boolean	false	Re-segment already-segmented blocks replacing previous segmentation
`provider`	string		AI provider id (anthropic
`rulesPath`	string		Path to a custom SRX 2.0 rules file (overrides the built-in ruleset)
`model`	string		Model name
`segmentSource`	boolean	true	Segment the source text
`treatIsolatedCodesAsWhitespace`	boolean	false	Treat isolated inline codes as whitespace during segmentation
`credential`	string		Stored credential name
`segmentTarget`	boolean	false	Segment existing target text
`trimLeadingWhitespace`	boolean	true	Exclude leading whitespace from each segment span
`instruction`	string		Optional guidance for how to chunk
`trimTrailingWhitespace`	boolean	true	Exclude trailing whitespace from each segment span
`maxChunkRunes`	integer		Soft upper bound on chunk size in characters (0 = model default)
`renumberCodes`	boolean	false	Renumber inline code IDs when materializing segments to a bilingual format

Configure it live

Configuration

Loading form…

Examples

Segment the source with the default rules

Mark sentence boundaries using the built-in rules.

segmentSource: true

Re-segment with a custom SRX file

Replace existing segmentation using project-specific rules.

engine: srx
overwriteSegmentation: true
rulesPath: ./rules/segmentation.srx

Processing notes

Operates on translatable blocks only; non-translatable blocks pass through unchanged.
When the source is written to the sentence layer, the segment count is recorded in the block property segment-count.

Limitations

The built-in rule set targets common Latin-script sentence boundaries; non-Latin scripts may need a custom SRX file.
The llm engine calls a hosted provider unless a local one is chosen, and needs a credential for a hosted provider.

← Back to the Tool Reference