Segment
The chunk a translator works on, usually one sentence or UI string, paired with its translated version in the translation tool.
What it is
A segment is the chunk a translation system presents as one job, usually a sentence in a document or one extracted UI string. The source segment is paired with a target segment, and that pair is what a TMS stores, reviews, counts, and tries to reuse later. A paragraph may become several segments, while a short button label is usually one by itself. Trados Studio and memoQ literally show them this way: source on the left, target on the right, one segment per row.
Segmentation happens before translation, using punctuation and format-specific rules, usually expressed as SRX rules. Keep a whole grammatical thought together and protect ICU MessageFormat placeholders and inline markup so a boundary cannot split them. Stable boundaries improve translation-memory reuse because old and new source chunks can be compared cleanly.
Gotcha: a segment is a workflow unit, not necessarily a complete sentence or a translation key. Split too aggressively and translators lose the grammar and context they need. Join too much and changing one clause makes the entire block look new, reducing useful matches and raising the word count sent for review.
Ask AI for it
Add a segmentation step before sending source text to the TMS. Use SRX sentence-boundary rules for prose, keep each extracted UI message as one segment, and treat ICU MessageFormat arguments, HTML inline elements, URLs, abbreviations, and decimal numbers as protected content that cannot be split. Give every segment a stable ID derived from its translation key and position, preserve source order, and add round-trip tests proving that joining the segments reproduces the original text exactly.