Key takeaways
- A translation memory stores pairs of source and target segments, usually sentences, and suggests them when the same or similar text appears again.
- Exact matches, fuzzy matches and internal repetitions are usually charged at lower rates than new words, so updates to existing documents cost less.
- TMX is the open exchange format that lets a translation memory move between tools and providers with little or no loss of data.
- A translation memory only saves money if it is kept clean; errors stored in it are repeated in every later project.
- Ask your provider to deliver your translation memory in TMX format, so you can reuse it if you change supplier.
This guide is for organisations that translate similar content repeatedly: product manuals, software releases, policies, contracts based on templates or e-commerce catalogues. It explains what a translation memory is, how it reduces the cost of each new project, and what buyers should ask about ownership and quality.
Translation memory is often confused with machine translation and glossaries, so this article also explains how the three differ.
What is a translation memory?
A translation memory (TM) is a database of previously translated text, stored as pairs of source and target segments. The European Commission's Joint Research Centre defines it this way on its DGT-Translation Memory page: “A translation memory is a collection of small text segments and their translations (referred to as translation units, TU).”
Translators use a TM inside a CAT tool (computer-assisted translation tool), software that splits a document into segments, usually sentences, and shows any stored translation for each one. The translator decides whether to accept, edit or reject the suggestion. Unlike machine translation, a TM only offers text that a human has already approved.
TMs can be very large. The European Commission's DGT-TM, built from EU legislation, is available in 24 languages and, according to the JRC, can be used to produce corpora for “any of the 276 language pairs or 552 language pair directions”.
How does translation memory reduce translation costs?
Translation memory reduces costs because text that has been translated before is usually charged at a lower rate than new text. Before quoting, the provider runs an analysis that compares your new document with the TM and sorts every segment into categories.
| Category | What it means | Usual effect on price |
|---|---|---|
| Context or in-context match | Identical sentence in the same surrounding text as before | Lowest rate; may still be checked |
| Exact (100%) match | Identical sentence found in the TM | Reduced rate for a check in context |
| Fuzzy match | Similar sentence, for example with one changed word or number | Sliding scale: the closer the match, the lower the rate |
| Repetition | Sentence that appears more than once in the new document | Reduced rate after the first occurrence |
| No match | New text | Full rate |
Worked example: updating a product manual
Imagine a 10,000-word manual that was translated last year and has now been revised. A hypothetical analysis might show 6,000 words as exact or context matches, 1,500 words as fuzzy matches and 2,500 words of new text. Only the 2,500 new words are charged at the full rate; the rest are charged at reduced rates for checking and editing. The same analysis also shows the translator exactly which sentences changed, which shortens the schedule as well as the bill.
The exact percentage bands and discounts differ between CAT tools and providers, so ask for the analysis report with every quote. For the other factors that set the price, see translation cost per word in the UK.
Which types of content benefit most from translation memory?
Translation memory saves most on content that is updated rather than rewritten. Typical high-saving content includes:
- new versions of user manuals and safety instructions;
- software and app strings released in regular updates;
- contracts and policies built from standard clauses;
- product catalogues with recurring specifications;
- e-learning courses that share modules.
Creative content, such as advertising and slogans, gains little because sentences rarely repeat. The TM still helps keep brand terms consistent.
Translation memory vs machine translation vs glossary: what is the difference?
A translation memory stores whole approved sentences, a glossary stores approved terms, and machine translation generates new text that nobody has approved yet. The three work together but are not interchangeable.
| Tool | What it contains | Approved by a human? | Main benefit |
|---|---|---|---|
| Translation memory | Sentence pairs from past projects | Yes | Reuse, lower cost, consistency |
| Glossary (termbase) | Key terms and their approved translations | Yes | Correct terminology |
| Machine translation | Text generated on demand | No, until post-edited | Speed on low-risk content |
Glossaries are covered in translation glossaries and style guides, and machine translation editing in what MTPE is.
Who owns the translation memory?
Ownership of a translation memory should be agreed in writing, because the TM represents translations you have paid for. Ask for it to be delivered in TMX (Translation Memory eXchange), the open XML format supported by mainstream CAT tools. The TMX 1.4b specification states that “The purpose of the TMX format is to provide a standard method to describe translation memory data that is being exchanged among tools and/or translation vendors, while introducing little or no loss of critical data during the process.”
A TM can contain confidential or personal data, because it holds every sentence you have translated. Treat it like the source documents: ask where it is stored, who can access it and how it will be deleted. Some machine translation services now offer TM features too; DeepL's privacy policy describes how TMs stored in a DeepL account are kept and deleted. Wider data handling is covered in data security when outsourcing translation.
Common mistakes with translation memory
A translation memory amplifies whatever goes into it, good or bad. These mistakes erode the savings:
- Editing translations after delivery without updating the TM. Your in-house corrections are lost and the old wording comes back in the next project.
- Letting several suppliers build separate TMs. Terminology splits and nobody has the full history.
- Accepting exact matches without context checks. The same English sentence may need different translations in different places.
- Never cleaning the TM. Outdated product names and superseded legal wording keep reappearing.
- Not asking for the analysis. Without it you cannot see whether matches were discounted.
For how TM fits into supplier selection, read how to choose a translation agency.
How TranslationWorks can help
TranslationWorks is a UK-based translation agency that provides professional human translation by native-speaking linguists in 150+ languages, for organisations that translate content repeatedly as well as one-off projects. If you already hold translation memories, glossaries or earlier translations, send them with your files so they can be taken into account in your free quote. TranslationWorks holds ISO 17100 (translation services) certification, ISO 27001 (information security) and ISO 9001 (quality management) registration. Explore our translation services and custom language services for ongoing programmes, or request a free quote within 24 hours.
