Glossary Updates6 new terms added to the glossaries · October 11, 2026, 13:38 CEST
AI TechDocKnowledge

Glossary · 6 · Translation and terminology

Translation memory

Also known as: TM, Translation memory database

German: Translation Memory

A translation memory is a database of aligned source and target segments from earlier translations, which computer-assisted translation tools search to offer stored translations of identical or similar segments (exact and fuzzy matches) for reuse; TMX (Translation Memory eXchange) is the XML format for exchanging such data between tools.

  • Translation and terminology
  • Standards
  • Technical documentation

In one sentence

A translation memory stores aligned source and target segments from earlier translations and offers them for reuse; TMX is its exchange format.

Example

After an update of a commissioning manual, 92 % of its segments are exact matches in the translation memory; translators work only on the changed and new segments.

How it applies

  • Consistency across versions: Reusing approved translations keeps wording consistent between product versions and documents and limits translation to what has changed. Content written in reusable units supports high match rates.
  • Exchange: TMX was developed by the OSCAR group of the Localization Industry Standards Association (LISA); version 1.4b is the last LISA version, and ETSI published it as ETSI GS LIS 002. Translation memories can thus move between tools and service providers.
  • Maintenance: A translation memory reproduces its errors. Corrections, especially in safety wording and terms, have to be made in the memory too, and outdated entries marked or removed; terms are checked against the termbase from terminology management.
  • Match is not release: An exact match was correct in its earlier context; whether it fits the new context is checked, and each language version is released as a whole.

Translation memory vs. machine translation

A translation memory returns translations that people produced and approved before. Machine translation generates new translations, which are then post-edited. Many tools combine both: they use memory matches first and machine translation for the rest. Translation memory data is also used to adapt machine translation engines, which makes its quality even more important.

External references