A segment is a single sentence, heading, or short phrase that a CAT tool treats as one translatable unit, stored and matched independently in translation memory.
Why Segment Matters
Segments matter because they define the unit that every part of the translation workflow operates on. A CAT editor — including Taia’s — splits a document into segments so a linguist can review one manageable piece at a time instead of scrolling through an entire file looking for changes — source text on one side, translation on the other, one segment highlighted at a time. Splitting at the sentence level also keeps context intact: a segment is usually long enough to carry meaning on its own, but short enough to compare quickly against the source.
Segmentation is also what makes translation memory possible. Each segment translated or edited is saved to TM paired with its source text, organized by language pair and surrounding context, so the exact same segment — or a fuzzy match close to it — can be reused the next time it appears. Without segment-level storage, TM would have to match entire documents, which almost never repeat exactly; matching at the segment level is what makes reuse practical at scale.
Segment status also drives project tracking: a segment can be untranslated, machine-translated, edited, or confirmed, and a project’s completion is measured as the share of segments in each state — not a rough word-count estimate.
A Concrete Example
A help-center article contains the sentence “Click Save to apply your changes.” That sentence is one segment. If the same instruction appears again later in the article, or in a different document translated the next month, translation memory recognizes it as a match to that same segment and suggests the saved translation instead of requiring the linguist to translate it again from scratch.
How Segment Works in Taia
In Taia’s CAT editor, every document is split into segments shown side by side — source on the left, translation on the right — and linguists move through them with keyboard shortcuts, confirming or editing one at a time. Every segment saved or corrected there is written automatically to translation memory, organized by language pair, so the same segment never has to be translated twice.