In order to create the translation
memory, the CAT
software divides the source text into segments. The segment usually corresponds
to a phrase, at least in running texts. Segmentation is governed by complex
rules based, in principle, on punctuation. See also translation
unit.