What is Translation Memory (TM)?

A translation memory is a database of source segments paired with their approved translations. When a new or changed string comes in, the TM is searched for identical or similar segments, and matches are offered to the translator or applied automatically. The goal is to never pay for, or manually retype, the same sentence twice.

Matches are scored by similarity. An exact match reproduces a previous translation verbatim; a fuzzy match is close but needs review, for example when one word changed in the source. TMs also enforce consistency, since recurring phrases like "Save changes" resolve to the same translation everywhere. Segmentation usually happens at the sentence level, which is what makes reuse across unrelated documents possible.

TM is different from machine translation: a TM only contains human-approved translations from your own past work, while MT generates new output. Mature workflows combine them, checking the TM first and falling back to MT for segments with no useful match.