How to migrate your translation memory to a new tool

TL;DR

• Your translation memory is yours. Any serious CAT tool, and any agency that has translated for you, can hand it over as a TMX file, the standard exchange format.
• A migration is three steps: export, clean, import. Exporting and importing take minutes. Cleaning is where the quality of your future matches is decided.
• The one thing most people miss is language codes. A memory tagged de instead of de-DE still imports, but it doesn’t say which German it is, and that changes how a new tool can use it.
• Comparing tools first? Our guide to the best translation memory software covers six of them.

---

What a translation memory migration actually moves

A translation memory (TM) is a database of sentence pairs: a source sentence and its approved translation, stored as one unit. Migrating it means taking those units out of one tool and putting them into another, so the next project starts from everything your team has already approved instead of from zero.

Most people don’t realize how much is in there. A company that has translated manuals or a website for a few years often has tens of thousands of approved sentences sitting in a tool, or with an agency, that nobody has opened since.

What a migration doesn’t move on its own is your terminology. Glossaries (termbases) are a separate file in almost every tool, so export them too.

Step 1: Export your translation memory as TMX

TMX (Translation Memory eXchange) is an XML format that every major CAT tool can read and write. Whatever tool you are leaving, TMX is the format to ask for.

Where to find it depends on where your memory lives:

• Trados Studio stores memories in its own .sdltm format. Open the Translation Memories view, select the memory and export it to TMX. Send the TMX, not the .sdltm file itself: it isn’t a standard format, and many tools can’t read it.
• memoQ exports a translation memory to TMX from its list of translation memories, in the Resource Console or inside a project.
• Phrase, Smartling and Crowdin each have an export option on the translation memory itself; choose TMX.
• Your old agency or freelancers. If someone else did your translations, the memory is usually in their tool, not yours. Ask them for a TMX export of every memory they used for you. It’s your content, and you own it.

Two habits that save trouble later:

1. Export everything before you cancel a subscription. Some tools lock or delete data when a plan ends.
2. Keep the original TMX files as an archive. Importing into a new tool can drop details the old one stored, such as who created an entry and when, so the original file is your record.

Step 2: Clean the TMX before you import it

This is the step people rush, and it decides how good your matches will be for years. A memory full of outdated or contradictory entries will suggest outdated or contradictory translations, faithfully, every time.

Check the language codes

Open the TMX in a text editor and look at the xml:lang attributes. You’ll see either a plain language (de, fr, es) or a regional variant (de-DE, de-AT, fr-CA).

A plain code doesn’t say which variant the translation is in. In Taia, for example, an entry tagged de is still offered as a suggestion in the editor and as an example for the AI, but Taia’s Auto-apply TM matches step skips it, and it doesn’t count toward the match analysis on a professional translation quote. If your memory really is German for Germany, tag it de-DE before you import it.

For Chinese and Serbian the variant is the script, so a plain zh or sr code isn’t used at all. Make sure those entries say zh-CN or zh-TW, sr-Latn or sr-Cyrl.

Remove what you wouldn’t approve today

• Old product names, retired slogans and translations from before a rebrand
• Entries that are just numbers, codes or a single tag
• Obvious mistakes your team has corrected since

Split it by domain

If one memory mixes legal contracts and marketing copy, consider splitting it. A sentence that’s right in a contract can be wrong in an ad. Separate memories let you choose what each project draws on, and in Taia a project can use several memories at once.

Make sure it’s UTF-8

TMX files are XML, and nearly all of them are UTF-8 already. If accented characters look broken when you open the file, re-export it from the old tool as UTF-8 rather than fixing it by hand.

Step 3: Import it into the new tool

In Taia, a translation memory doesn’t need its languages set in advance:

1. Create a translation memory (name, an optional description, and whether it belongs to the whole organization or specific teams).
2. Open it, choose Import TMX from its menu and upload one or more .tmx files.
3. The import runs in the background, and you get a notification when it’s done.

What happens to your entries on the way in:

• Every language pair in the file is imported. A TMX with English to German, French and Italian arrives as all three.
• Exact duplicates are skipped, so importing the same file twice doesn’t double your memory.
• Different translations of the same sentence are both kept, so you can review which one should stay.
• Imported entries are labeled “TMX Import”, so reviewers can see where a match came from, and you can filter the memory by it.

The full steps, with screenshots, are in the help center: Import a TMX file.

Translation memory is available on Pro plans and above.

Step 4: Check it worked

Don’t take the notification’s word for it. Open a recent file that your old tool would have matched heavily, and look at the matches.

• Do the high matches show up where you expect them?
• Are the right language variants being applied, not just suggested?
• Are there two competing translations for your most common sentences? Pick one and delete the other.

Then attach the memory to your next real project. From there, every sentence your team confirms in the translation editor is saved back into it, so the memory keeps growing from where your old tool left off.

What a migrated memory does for you

On professional translation, it lowers the price. Matches and repetitions are billed at a discounted per-word rate, and the higher the match, the lower the rate.

On AI translation, it changes the result, not the credits. Credits are charged per word either way. What the memory does is steer the first draft toward the wording you’ve already approved: with Auto-apply TM matches switched on, 100% matches go straight into the draft, and on Advanced and Premium AI similar sentences are handed to the AI as examples. So there’s less to correct at review.

Frequently Asked Questions

What file format should I export my translation memory in?

TMX (Translation Memory eXchange). It’s the industry-standard format for moving translation memories between tools, and every major CAT tool can export and import it. Native formats such as Trados’s .sdltm aren’t a standard, so many tools can’t read them.

Can I get my translation memory from my old translation agency?

Yes. A translation memory built from your content is yours, and any agency that used a CAT tool for your projects can export it as TMX. Ask for every memory they used for your account, in all your language pairs.

Will I lose anything when I migrate a translation memory?

The sentence pairs move intact. Details some tools store alongside them, such as the original author, creation date or notes, may not carry over, so keep the original TMX files as an archive.

Does Taia import multilingual TMX files?

Yes. Every language pair in the file is imported into the translation memory, exact duplicates are skipped, and each entry is labeled “TMX Import” so you can see where it came from.

Can I export my translation memory from Taia again later?

Yes. Export TMX downloads the whole memory, every language pair and every entry, as a TMX file whenever you want it. There’s no lock-in.

Do I need to migrate my glossary separately?

Yes. Glossaries are separate from translation memories in almost every tool. Taia imports glossaries from Excel (XLSX) files using the glossary’s own template; see Import an existing glossary.

Bring your translation memory to Taia