What Is Translation Memory and Why Agencies Should Care
7 min read
July 2026
Mulyadi Subali
CAT Tools · Agency Resources
Every project manager who has sent a file for a quote has seen the phrase "TM leverage applied." Fewer have had someone explain what it actually means for their budget, their timeline, or the translator receiving the file. I want to walk through translation memory the way I explain it to new clients — not as a marketing term, but as the piece of infrastructure that makes long-term agency relationships cheaper and more consistent over time.
What Is Translation Memory?
Translation Memory (TM)
A database that stores every sentence or segment I have translated, paired with its source. When a new document comes in, my CAT tool checks each segment against the TM and tells me if something similar — or identical — has been translated before.
What TM Is Not
TM is not machine translation. It does not generate new text. It only recalls text a human has already translated and approved. If a segment has never appeared before, TM offers nothing — the translation still has to be done from scratch.
The confusion between TM and MT is common enough that it is worth stating plainly: translation memory is a memory of human work, not a generator of new output. That distinction matters when a client asks why a "100% match" still needs a translator to review it — reused text still has to be checked against the new document's context.
How TM Leverage Works in Practice
1. Segmentation
The CAT tool breaks the source document into segments — usually sentences, sometimes list items or table cells — and compares each one against the TM database.
2. Match scoring
Each segment gets a percentage score. 100% means an exact match to something already in the TM. Anything below that is a "fuzzy match" — close, but requiring editing.
3. Editing or translating
High matches get reviewed and adjusted for context. Low matches and no-matches get translated normally. Either way, the new work is added back into the TM for future projects.
4. Delivery and TM update
The finished file goes back to the agency, and — depending on the setup — the updated TM either stays on my machine, syncs to the agency's server, or gets delivered alongside the file for the client's own archive.
Why This Matters to a Project Manager Specifically
Lower cost over time. Fuzzy and exact matches are typically billed at a reduced rate compared to new words — the client benefits from every prior project in the same TM.
Consistency across large or recurring accounts. If a client has used a specific phrasing for a legal clause or a product name for three years, TM surfaces it automatically instead of relying on a translator's memory or a style guide lookup.
Faster turnaround on repeat content. Manuals, contracts with boilerplate clauses, and recurring marketing formats all benefit — the more repetitive the content type, the more TM leverage helps.
An asset that belongs to the client, not the vendor. A well-maintained TM is something an agency can hand to a new translator without losing years of terminology decisions.
None of this replaces human judgment. A 95% fuzzy match on a legal document still needs the same scrutiny as a new sentence — the 5% difference could be a negated clause or a changed number, and those are exactly the kind of changes that matter most.
TM Formats and Compatibility
Translation memories are portable between tools when exported correctly, which matters if an agency works with multiple translators on different CAT tools.
TMX
The open standard format. Exports from Trados, memoQ, Wordfast, and most other CAT tools, and imports cleanly across all of them. This is the format to request if you are not sure what the receiving translator uses.
SDLTM
Trados Studio's native database format. Works directly inside Trados without conversion, but needs to be exported to TMX first for use in other tools.
MQXLIFF-linked TM
memoQ's server-based memories, often used in cloud/TMS workflows where the TM lives centrally and translators connect to it rather than working from a local file.
A Common Misconception Worth Correcting
"If it's a 100% match, it shouldn't need review."
A 100% match means the source sentence is identical to one seen before — it does not confirm that the sentence fits correctly in its new surrounding context, or that a target term hasn't been intentionally updated since the last project (a rebrand, a legal amendment, a changed product name). I still review 100% matches in context; I just do it faster than a from-scratch segment.
Building and Maintaining a TM Over a Long-Term Relationship
For agencies I have worked with over multiple years, the TM becomes a shared asset rather than a personal tool. I keep client-specific TMs separate from general-domain ones, apply glossaries alongside the TM so terminology stays locked even when a segment is new, and flag inconsistencies back to the project manager rather than silently overriding an established term. This is the same discipline I applied building the in-house terminology base during my time as Language Lead on the Toyota account — a large TM is only as useful as its consistency, and consistency has to be actively maintained, not assumed.
Working with a TM-Ready Indonesian Translator
CAT tool and TM-compatible for your production pipeline.
Trados Studio, memoQ, Phrase, XTM, and Smartcat — direct TMS access supported, TMX import/export handled cleanly, and existing glossaries applied from segment one. Send documents for a quote and I'll confirm TM/format compatibility before starting.
No. Translation memory only recalls segments that have already been translated by a human and approved. Any new sentence — anything that hasn't appeared in a prior project — still requires full translation. TM speeds up repeat and near-repeat content; it does not generate new translations on its own.
A fuzzy match is a segment in the new document that closely resembles, but does not exactly match, a segment already stored in the TM. CAT tools score these as a percentage — for example, an 85% match — and the translator edits the recalled text to fit the new context rather than translating from scratch.
This depends on the agreement, but standard practice is that the TM belongs to the paying client, since it was built from content the client commissioned and paid for. Agencies typically deliver the updated TM alongside the final files, or maintain it on a shared server the client can access.
No, and this is one of the most common points of confusion for people outside the industry. Translation memory recalls existing human-translated segments. Machine translation generates new text using an algorithm, with no guarantee it has ever been reviewed by a person. The two are sometimes used together — in an MTPE workflow, for example — but they solve different problems.
Most professional translators and agencies apply a reduced rate for TM matches, scaled by match percentage — exact matches priced lowest, fuzzy matches on a sliding scale, and new words priced at the standard rate. This rewards clients for building up a TM over repeat projects, since each subsequent project draws on a larger memory. Exact discount structures vary by translator and are typically confirmed at the quote stage rather than fixed universally.
TMX is the safest choice for compatibility, since it is the open, tool-agnostic standard that imports cleanly into Trados, memoQ, Wordfast, and most other CAT tools. If your agency's TM lives natively in a specific tool's format — SDLTM for Trados, for instance — ask whether the receiving translator's tool can import it directly, or request a TMX export to avoid conversion issues.