Translation technology has evolved to levels that once seemed impossible — and translation memory sits at the center of it.
Alongside the rise of AI and machine translation, tools such as termbases, glossaries, and translation memories have become the right hand of today’s translators.
Translation memory has been around since the 1970s and has grown into a powerful resource for translators and translation agencies aiming for scalability, quality, and speed. In this guide, we explain what a translation memory is, how it works, how it differs from machine translation, how it compares to glossaries and CAT tools, and why it matters for your projects.
Quick answer: A translation memory (TM) is a database that stores previously translated text segments — called translation units — in source-and-target language pairs. When new content is translated, the TM finds matching segments so linguists reuse approved human translations instead of starting from scratch, which lowers cost, speeds delivery, and keeps terminology consistent.
What Is a Translation Memory?
A translation memory is a feature of computer-assisted translation (CAT) tools that stores text segments in the source and target languages — from strings of words and sentences to whole paragraphs.
These segments are called translation units and can be retrieved automatically when a new document is uploaded for translation. This lets professional linguists leverage existing, human-verified translations instead of starting from scratch.
One of the greatest benefits of translation memory is that it significantly accelerates the translation process by avoiding redundant effort. It also keeps terminology and style consistent — not only within a single project, but across every project stored in the memory database.
Why Should You Care About Translation Memory as a Client?
In professional translation agencies, the translation memory grows with every project, so the chances of finding reusable matches increase over time. This reduces your per-word translation cost by discounting repetitive segments — and it means you receive work with higher consistency, accuracy, and quality on every delivery.
Key Benefits of Translation Memory
For businesses running ongoing or multilingual translation, a translation memory delivers four compounding advantages:
- Lower cost: Repeated and similar segments are reused, so you pay reduced rates — or nothing — for 100% and fuzzy matches. Savings grow as the memory grows.
- Faster turnaround: Linguists translate each segment once, so large and recurring projects move faster.
- Consistency: Approved wording and terminology are reused across every document, protecting your brand and technical accuracy.
- Scalability: As volumes rise, the memory does the heavy lifting — ideal for website and software localization with frequent updates.
Translation Memory vs. Machine Translation: What Is the Difference?
Translation memory (TM) and machine translation (MT) both help produce translations, but how they operate is what sets them apart. A translation memory stores previously translated, human-approved segments and reuses them; machine translation uses algorithms to generate entirely new translations automatically.
| Aspect | Translation Memory (TM) | Machine Translation (MT) |
|---|---|---|
| How it works | Reuses human-translated segments from a database | Generates new translations by algorithm |
| Human involvement | Linguist leverages and approves every match | Automatic; no human unless post-edited |
| Quality | Consistent and pre-verified | Variable; usually needs post-editing |
| Best for | Repetitive, recurring, consistent content | Fast first drafts and large volumes |
| Consistency | High — approved terminology is reused | Depends on the engine and content |
How Translation Memories Work
Here are the steps a translator follows when working with a translation memory inside a CAT tool or translation management system (TMS):
- The translator uploads the text that needs translating into a CAT tool or TMS.
- The translation memory identifies matching content between the source text and previously translated segments.
- Matching segments are grouped into categories based on how similar they are.
- The TM suggests translations for the translator to choose from.
- The translator accepts the suggestion as-is, edits it, or writes a new translation.
The Different Types of Translation Memory Matches
Because translations are not always identical, the TM categorizes content by similarity. Each match type carries a different editing effort — and a different cost impact.
| Match type | Similarity | Editing needed | Cost impact |
|---|---|---|---|
| Perfect / exact match | 100% | None | Lowest (highest discount) |
| Nearly exact match | 95–99% | Minor (numbers, punctuation, spacing) | Low |
| High fuzzy match | 85–95% | Small edits | Reduced |
| Medium fuzzy match | 75–85% | Moderate edits | Partial |
| Low fuzzy match | Below 75% | Substantial edits or retranslation | Little to none |
| No match | 0% | Full translation | Full rate |
The TM may also suggest terminology matches when individual words or phrases correspond to entries in a connected terminology database.
Translation Memory vs. Glossary vs. Terminology Base
These three assets are often confused, but they do different jobs and work best together. A translation memory stores whole translated segments (sentences and paragraphs). A glossary or terminology base stores approved terms — product names, technical vocabulary, and preferred wording — with their target-language equivalents. The TM speeds up and reuses full sentences; the termbase keeps individual words on-brand and accurate across every project.
Translation Memory vs. CAT Tools vs. TMS
A translation memory is not a standalone product — it lives inside larger systems. A CAT tool is the linguist’s workspace, where the TM, termbase, and quality checks come together. A translation management system (TMS) sits above that, orchestrating files, workflows, vendors, and the shared translation memory across an organization. If you are comparing platforms, see our guide to CAT tools vs. translation management systems and the wider landscape of localization tools.
Who Owns Your Translation Memory?
Your translation memory is a valuable linguistic asset — and ownership matters. A reputable language service provider builds and maintains a dedicated TM for your account, keeps it secure, and returns it on request, so the value you invest in translation stays yours. Before you commit to a partner, ask who owns the TM, how it is stored and protected, and whether you can export it. Clear ownership prevents vendor lock-in and protects your terminology over the long term.
Translation Memory and Machine Translation, Working Together
Modern workflows do not choose between TM and MT — they combine them. The translation memory handles repeated and previously approved segments, machine translation drafts the remaining new content, and human linguists review and refine the result through machine translation post-editing (MTPE). This balance delivers speed and cost efficiency without giving up the consistency and accuracy that only human-verified translation provides. Translation memories are also portable: the industry-standard TMX (Translation Memory eXchange) format lets you move your memory between tools and providers.
Final Thoughts
Translation memory speeds up workflows and keeps quality consistent — but to get the full benefit, it needs to be built, maintained, and applied by experienced language specialists. Used well, it turns every project you complete into savings and consistency on the next one.
At AsiaLocalize, our ISO-certified linguists pair advanced translation memory systems and CAT technology with tightly controlled translation processes to streamline your document translation and localization projects — so you receive consistent, on-brand deliverables on time and on budget, in 120+ languages. Request a quote to see how a dedicated translation memory can cut your costs.
Translation Memory FAQ
Is translation memory the same as machine translation?
No. A translation memory reuses human-translated segments stored in a database, while machine translation generates new translations automatically. TM ensures consistency and pre-verified quality; MT provides speed but usually needs human review.
Does translation memory reduce translation costs?
Yes. Repeated and similar segments are reused, so you pay reduced rates — or nothing — for 100% and fuzzy matches. The more you translate, the larger the memory grows and the more you save.
What is a fuzzy match in translation memory?
A fuzzy match is a segment that partially matches a stored translation, usually between 75% and 99% similar. The translator adjusts the suggestion rather than translating from scratch, with the cost depending on how close the match is.
Who owns the translation memory?
Ideally, the client owns their translation memory. A reputable language provider builds and maintains a dedicated TM for you, keeps it secure, and returns it on request, so your linguistic assets remain yours.
What is the difference between a translation memory and a glossary?
A translation memory stores whole translated segments (sentences and paragraphs), while a glossary or terminology base stores approved individual terms and their translations. They work together to keep both wording and terminology consistent.
Can translation memory work with machine translation?
Yes. Modern workflows combine translation memory, machine translation, and human post-editing (MTPE): the TM handles repeated segments, MT drafts new content, and linguists review the output — balancing speed, cost, and quality.
