The loop
Localize → Review → Finalize → Translation Memory → Future localization
Translation Memory is approved language memory. It is not model training — no AI model is retrained from your content.
How it updates
How it updates: Translation Memory is the one context type that grows from approved work — approved source and target pairs from finalized localization, plus manual entries. Each entry keeps its source language, target language, approved translation, provenance and version history, with one current approved version.
Why it matters
Not every sentence needs to be translated from scratch.
When Localizethat recognizes language that has already been approved, previous translation decisions can provide useful context for the new localization.
How entries get there
Finalized translations can be added automatically to Translation Memory. When you finalize a target language with “Save approved translations to Translation Memory” switched on, every approved source and target pair becomes reusable company context, and you see exactly how many were added, updated or already known.
You can still add entries manually in Company context → Translation Memory, and every entry shows whether it came from a finalized localization or was added by hand.
Example
PREVIOUSLY APPROVED SOURCE
External reviewers are free.
PREVIOUSLY APPROVED SWEDISH
Externa granskare är kostnadsfria.
NEW SOURCE
External reviewers are free and do not need a Localizethat account.
PREVIOUS APPROVED MATCH
“External reviewers are free.”
→
“Externa granskare är kostnadsfria.”
STILL REQUIRES LOCALIZATION
“and do not need a Localizethat account.”
The previous decision can be reused while the new part is still localized in context.
Important distinction
Translation Memory should preserve useful previous translation decisions. It should not force an old translation when the new context requires different wording.