200,000 segments after five years of work - sounds like a goldmine. But then you open your TM and see: three different translations of “Arbeitsvertrag” (because in 2022 you went with “employment contract”, in 2023 - “labor agreement”, and in 2024 you switched back to “employment contract”), dozens of duplicates, outdated company names, and segments from a project you don’t even remember. Sound familiar? Then this article is for you.
Translation Memory (TM) is probably a translator’s most valuable asset after their own brain. But only if you maintain it properly, clean it regularly, and understand how to turn years of work into a competitive edge. Let’s break down how to do that.
What Is Translation Memory and Why It’s Your Top Asset¶
If you already work with CAT tools, you know what TM is. But just in case - here’s a quick refresher.
Translation Memory is a database that stores pairs of “source + translation” at the sentence (segment) level. Every time you translate a new segment, your CAT tool saves it to the TM. Next time a matching or similar sentence comes up, the tool automatically suggests the previous translation.
The standard storage format is TMX (Translation Memory eXchange). It’s an XML file that every CAT tool understands: Trados, memoQ, Smartcat, OmegaT - you name it. You can freely move your TM between different tools without losing data.
Why is TM a goldmine? Here are real numbers. According to Taia, companies using TM save 30-50% on translations after the first 6 months. Exact matches can cut costs by 70-90%, fuzzy matches by 40-60%. One SaaS company saved $45,000 per year by reusing 12,000 segments from their TM.
For you as a translator, this means: the larger and cleaner your TM, the faster you work and the more you earn per hour of actual time.
How to Organize Your Translation Memory Properly¶
The worst thing you can do is dump everything into a single TM. Legal translations, medical, marketing - all in one file. Within a year you’ll have a mess where “Behandlung” gets suggested as both “treatment” (from a medical project) and “processing” (from a technical one).
TM Structure: Divide and Conquer¶
Here’s a working structure that experienced translators use:
| Level | Example | Purpose |
|---|---|---|
| By domain | TM_Legal_DE-EN, TM_Medical_DE-EN | Different terminology doesn’t mix |
| By client | TM_Siemens_DE-EN, TM_Bosch_DE-EN | Each client has their own terms and style |
| By language pair | TM_DE-EN, TM_FR-EN | Obvious, but many people skip this |
| Main + reference | Main TM + Reference TM | Main for active projects, reference for old translations as context |
One rule that saves you from chaos: main TM for active projects (new, verified translations) + reference TM for everything old (to peek at, but not auto-insert).
Metadata Is Your Friend¶
Every segment in a TM can carry metadata: creation date, translator name, project name, client. Fill these in. When you need to decide which of three translation variants is correct two years from now - metadata will tell you who translated it, when, and for whom.
memoQ has a great feature - you can assign quality ratings to segments. In Trados you can filter by date, author, and project. Use these capabilities.
Cleaning Your TM: How and Why¶
One translator on the Trusted Translations forum described TM cleanup as a “Titanic task.” And that’s true - if you let things slide for years. But if you do it regularly, it’s 2-3 hours per quarter.
What Needs Cleaning¶
Duplicates with different translations. The same source segment with two or three different translations. Your CAT tool doesn’t know which to pick, you waste time choosing manually. Solution: keep the correct variant, delete the rest.
Outdated terminology. A client company renamed their product. Legislation changed and a term got updated. You refined your glossary. But the TM still has old translations lurking as fuzzy matches.
Empty or “junk” segments. Numbers, dates, single words, one-character strings - all of this ends up in your TM and clutters it. A period segment “.” with a translation of “.” definitely won’t help you.
Segments with formatting errors. Broken tags, extra spaces, wrong formatting - all of this degrades auto-substitution quality.
Tools for Cleaning¶
| Tool | Price | What it does |
|---|---|---|
| Trados TM Maintenance | Included in Trados | Duplicate removal, filtering, batch editing |
| memoQ TM Editor | Included in memoQ | Editing, filters, export/import |
| Olifant (Okapi) | Free | Viewing, editing and filtering TMX files |
| ApSIC Xbench | €99/year (v2.9 free) | QA reports, inconsistency search, terminology checks |
Here’s a professional cleanup workflow:
- Export your TM to TMX format
- Open it in Xbench or Olifant - get a report on duplicates and inconsistencies
- Go through the report, make a decision on each problematic segment
- Delete or edit the problematic segments
- Import the cleaned TM back into your working CAT
How Often to Clean¶
At minimum - once per quarter. Ideally - after every large project. As one author on POEditor put it: “TM cleanup isn’t a one-time project but the beginning of a process. Without regular maintenance, you’ll be back to square one within months.”
Set a calendar reminder. Seriously. 2-3 hours every 3 months will save you dozens of hours throughout the year.
How to Monetize Your Translation Memory¶
You’ve been translating legal documents DE>EN for five years. You have 150,000 segments with verified legal terminology. That has value. The question is - how to capture it.
Option 1: Offer Discounts via TM Leverage¶
When a new client brings text that matches 60% of your TM - you can do the job twice as fast. Instead of quietly earning more per hour, offer the client a discount. This is standard industry practice:
| Match type | Discount from full rate |
|---|---|
| Exact match (100%) | 70-90% discount |
| Fuzzy match (75-99%) | 30-60% discount |
| Repetitions (repeats within text) | 70-90% discount |
| No match (new segments) | Full rate |
Yes, you earn less per word - but you win on speed and client loyalty. The client sees savings, comes back again, and your TM keeps growing.
Option 2: TM as a Competitive Advantage¶
“I already have 100,000 segments of legal translations in the DE>EN pair” - that sentence in your ProZ profile or in a reply to a client says more than any certificate. It’s proof of experience and a guarantee of quality and consistency.
When looking for new clients, mention your TM. Agencies understand and value this - lower costs for them, higher quality for the end client.
Option 3: Selling TM and Glossaries¶
Platforms exist for this: TM-Town (glossary sales, 20% commission), TTMEM.com (TM Stock - a marketplace for TM). Sounds appealing, but there are serious caveats.
Translator and blogger Pieter Beens wrote an honest take on this: “The most important disadvantage is that you can scramble your knowledge and expertise which took years to build to make some pennies while putting yourself at some risk.” And further: “If you sell only generic segments without specific terminology - the TM becomes useless to the buyer. But if you sell it with terminology - you lose your competitive edge.”
My advice: don’t sell your TM directly. You’re better off using it as a competitive advantage through discounts (option 1) and as proof of expertise (option 2).
Who Owns Translation Memory: Know Your Rights¶
This is a question many translators ignore - and they shouldn’t. Who owns the TM that you fill while working for a client or agency?
The Legal Side¶
As a general rule, if you’re a freelancer - copyright on the translation belongs to you, unless the contract says otherwise. But many agencies include clauses about TM rights transfer. Some require you to hand over the TM after a project ends.
One ProZ user wrote: “Translation Memory is the translator’s intellectual property.” And legally that’s true - unless the contract states otherwise.
What to Do¶
- Always read the contract, especially clauses about IP (intellectual property) and TM ownership
- If an agency demands the TM - discuss it before work begins, not after
- Keep a parallel personal TM for general translations (not client-specific ones)
- If you’re drafting a contract with a client, clearly state who keeps the TM after the project ends
The recommendation from oneword.de (a German translation agency) is straightforward: “It is advisable to settle the question of utilization and usage rights before commissioning translation services, and to ensure contractually that you have the sole right of use to your translations and TM files.”
Healthy TM Checklist¶
Here’s a quick checklist you can pin to your monitor:
- TMs are separated by domain and language pair
- After each project - quick review of new segments
- Once per quarter - full cleanup (duplicates, outdated terms)
- Metadata is filled in (date, author, project)
- TM backup is stored separately (cloud storage or external drive)
- Contracts checked for TM ownership clauses
Honestly, most translators don’t do even half of this. But those who do - they work faster, earn more, and have far less headache with translation consistency. Your call.
FAQ¶
How many segments does a TM need before it starts being useful?¶
Even 1,000 segments make a difference - especially if you work in a single niche. The real impact kicks in at 10,000+ segments, when fuzzy matches start appearing in almost every new text. But segment quality matters more than quantity - 5,000 clean, verified segments are worth more than 50,000 cluttered ones.
Can I merge multiple TMs into one?¶
Yes, all CAT tools support merging TM files. But before merging, make sure to clean each TM separately - otherwise you’re just multiplying chaos. In Trados you do this via Upgrade Translation Memories, in memoQ through Import. The TMX format guarantees compatibility.
How do I transfer a TM from one CAT tool to another?¶
Export your TM in TMX format - it’s the standard supported by Trados, memoQ, OmegaT, Smartcat, Wordfast, and every other CAT tool. Import the TMX file into your new tool. Some metadata or tool-specific attributes might not carry over, but the actual translations will transfer 100%.
What should I do with my TM after switching specializations?¶
Don’t delete old TMs. Move them to a “reference” category - they won’t auto-insert into new projects, but you can search through them manually if needed. Who knows - maybe an old client will come back a year later with updates to those same texts.
Do I need to back up my Translation Memory?¶
Absolutely. Your TM is the result of years of work, and losing it to a disk failure or virus is a disaster. Keep a copy in cloud storage (Google Drive, Dropbox) and on an external drive. Update your backup once a month. TMX files are small - even a TM with 500,000 segments weighs less than 100 MB.