Technology ยท Core concept
Translation Memory
A translation memory (TM) stores every translated segment alongside its source. When similar text appears again, the TM offers the previous translation for reuse, improving consistency and reducing cost and turnaround.
Also known as: translation memory, tm, translation memories, tm match
What the TM does with three incoming segments
| Incoming segment | TM result |
|---|---|
| Checkout failed. Please try again. | |
| Checkout failed. Try again later. | |
| Your session expired. |
What it is
A database of segment pairs. Each row holds one source segment and its approved translation, plus metadata: language pair, domain, client, date, and who translated it. Segments are usually sentences or UI strings, not paragraphs.
How matching works
When new text arrives, the TM compares each segment with what it stores. An exact match (100%) reuses the stored translation. A fuzzy match (for example 85%) suggests the previous translation with the differences highlighted. Anything below the fuzzy threshold is treated as new and translated from scratch.
Why leverage is not free
Reusing a 100% match still needs review. Context changes, and a segment that was correct last year can be wrong in a new interface. Professional workflows treat matches as suggestions with provenance, never as finished output.
Common mistakes
Paying for 100% matches as if they were new translations. Importing unreviewed machine-translated output into a TM, which poisons every future match that reads from it. Sharing one TM across unrelated clients, which dilutes the context the matches were correct in.