Answer
TMX Files Explained: Structure, Uses, and Best Practices
TMX files are XML translation memory exchange files that move bilingual segments between CAT tools. Learn the structure, elements, and TMX 1.4 best practices.
Translation memory interchange rides on one open XML format that every production CAT tool reads and writes without data loss. This page walks the TMX 1.4b element tree the way Trados, memoQ, Phrase, OmegaT, Wordfast, Smartcat, MateCat, and Across actually parse it — then shows how to leverage, export, and validate a TMX without poisoning downstream fuzzy matches. TMX sits beside XLIFF in the open interchange stack; the live bilingual-file side is covered XLIFF: The XML Localization Interchange Format Translators Actually Ship.
Key points
- TMX — short for Translation Memory eXchange — is an open XML standard whose current stable version, 1.4b, was finalized in 2005 and remains the universal format for moving translation memories between CAT tools without data loss.
- Every TMX file is UTF-8 encoded and validated against the TMX 1.4 DTD; the root element is
<tmx version="1.4">, which contains exactly one<header>and one<body>. - The
<header>carries seven required attributes —creationtool,creationtoolversion,segtype,o-tmf,adminlang,srclang, anddatatype— that tell any receiving tool exactly how the memory was built and segmented. - Each
<tu>(translation unit) holds one logical segment; inside it, one<tuv>per language variant carries the translated text wrapped in a<seg>element, withxml:langidentifying the locale (e.g.,en-US,fr-FR). - TMX 1.4b achieves universal round-trip interoperability across all major CAT environments; the never-ratified 2.0 draft is not in production use and should not be targeted for export.
- TMX leverage never bypasses an LQA quality gate: a 100 % match still requires review against an MQM-style scorecard aligned to ISO 17100 before delivery on a certified lane.
- For compatibility across tools, receivers silently ignore unknown elements and attributes rather than hard-failing on import — a design principle baked into how the standard was conceived.
What is a TMX file?
A TMX file is an open XML document, standardized as Translation Memory eXchange 1.4b, that stores aligned source-and-target translation segments so a translation memory can move between CAT tools — Trados, memoQ, Phrase, OmegaT, Wordfast — without data loss. The file is UTF-8 encoded and validates against the TMX 1.4 DTD. Each stored segment pair is reusable across projects, tools, and clients.
Two unrelated file formats share the .tmx extension. Translators who land here looking for a game tilemap need a different reference. The disambiguation table below resolves the confusion that caused the search.
| Term | What it is | When you need it |
|---|---|---|
| Translation Memory eXchange (.tmx) | Open XML format for bilingual/multilingual translation memory | Moving TM between Trados, memoQ, Phrase, OmegaT, Wordfast |
| Tiled Map XML (.tmx) | XML tilemap format used by the Tiled game-map editor | Building 2D game levels in Unity, Godot, or a custom engine |
| XLIFF (.xlf /.xliff) | OASIS XML bilingual-file format for one project in flight | Handing a live translation file between CAT tools or a TMS |
| TBX (.tbx) | ISO XML term-base exchange format for glossary entries | Sharing term bases with definitions and part-of-speech |
What does TMX stand for, and who standardized it?
TMX stands for Translation Memory eXchange. The LISA OSCAR group ratified version 1.4b in 2005. It remains the de facto open interchange standard for translation memories across every production-grade CAT tool. LISA dissolved in 2011, leaving 1.4b as the final stable specification the industry writes and reads today.
Translation Memory eXchange vs Tiled Map XML: which.tmx did you mean?
Two unrelated formats share the .tmx extension: Translation Memory eXchange (localization) and Tiled Map XML (game-development tilemaps). This page covers the localization format. Readers landing here for the Tiled game-map.tmx need the Mapeditor.org reference instead — the two formats share no elements, no schema, and no tooling.
What is the structure of a TMX file?
A TMX file's structure is an XML tree with two top-level children under the root <tmx> element: a <header> carrying metadata and a <body> containing translation units. Each <tu> holds one or more <tuv> language variants. Each <tuv> wraps its text in a single <seg> element.
The element tree drawn as a flowchart:
flowchart TD A[XML declaration UTF-8] --> B["<tmx version=1.4>"] B --> C["<header> metadata"] B --> D["<body> translation units"] D --> E["<tu> one logical segment"] E --> F["<tuv xml:lang=en-US>"] E --> G["<tuv xml:lang=es-ES>"] F --> H["<seg> source text"] G --> I["<seg> target text"] E --> J["<prop> typed metadata"] E --> K["<note> comment"]
What goes in the XML declaration and <tmx> root element?
Every TMX file opens with an XML declaration specifying UTF-8 encoding, followed by the root <tmx> element whose version attribute reads "1.4". The root element contains exactly one <header> and exactly one <body>. A DOCTYPE reference points at the TMX 1.4 DTD for validation.
What attributes does the <header> element carry?
The <header> element carries 8 required attributes that describe the memory's provenance. Optional attributes add timestamps and change tracking. The table below splits them for a translator writing or auditing a header by hand.
| Attribute | Required? | Role |
|---|---|---|
creationtool | Required | Name of the tool that created the TMX (e.g. "SDL Trados Studio") |
creationtoolversion | Required | Version string of the creation tool |
segtype | Required | Segmentation granularity: block, paragraph, sentence, or phrase |
o-tmf | Required | Original translation-memory format before export |
adminlang | Required | Administrative language for notes/props (e.g. "en-US") |
srclang | Required | Source language of the memory, or "*all*" for multidirectional |
datatype | Required | Content type: plaintext, html, xml, winres, rtf, etc. |
o-encoding | Optional | Original character encoding before UTF-8 conversion |
creationdate | Optional | UTC timestamp in YYYYMMDDThhmmssZ format |
creationid | Optional | User or process that created the memory |
changedate | Optional | UTC timestamp of last modification |
changeid | Optional | User or process that last modified the memory |
What is a <tu> translation unit, and what does each <tuv> variant contain?
A <tu> element is one translation unit — one logical segment of source text and its translations. Each <tu> contains one or more <tuv> variants identified by an xml:lang attribute. Each <tuv> wraps its translated text inside a single <seg> element. A bilingual TMX has two <tuv> per <tu>; a multilingual TMX has three or more.
How do <prop> and <note> elements add context to a translation unit?
The <prop> element attaches a typed property to a <tu> or <tuv>. Common property types include project, client, domain, and status. The <note> element attaches a free-text comment. Both are optional but critical for translators filtering a leveraged memory by domain or project.
<prop type="project">ACME-2024-Q4</prop>— filters matches by delivery batch.<prop type="domain">legal</prop>— separates legal and medical TM pools.<prop type="client">ClientX</prop>— scopes leverage to one buyer's style.<prop type="status">approved</prop>— excludes draft translations from fuzzy results.<note>Deprecated term — use "login" instead of "log on"</note>— carries human context.
How does TMX mark up inline formatting inside a segment?
TMX uses 6 inline content elements — <bpt>, <ept>, <ph>, <it>, <hi>, and <sub> — to mark up paired tags, standalone placeholders, isolated tags, highlighted spans, and sub-flow text. The i attribute pairs a <bpt> open with its matching <ept> close. The x attribute matches identical codes across <tuv> variants so a receiving tool keeps formatting aligned.
| Element | Stands for | What it marks | Key attributes |
|---|---|---|---|
<bpt> | Begin paired tag | Opening half of a paired native tag (e.g. <b>) | i, x, type |
<ept> | End paired tag | Closing half matched to a <bpt> by i | i |
<ph> | Placeholder | Standalone native code: variable, line break, image | x, type, assoc |
<it> | Isolated tag | Opening or closing tag without its pair in this segment | pos, x, type |
<hi> | Highlight | Span of text flagged for the translator's attention | x, type |
<sub> | Sub-flow | Nested translatable text inside a tag (e.g. alt text) | datatype, type |
What does a working TMX file example actually look like?
A working TMX file example contains the XML declaration, a <tmx version="1.4"> root, a <header> block with creationtool and srclang, and a <body> holding one <tu> with two <tuv> variants — one English source, one target language — each wrapping text in a <seg>. The two examples below are specimens; replace placeholders with real values.
Minimal bilingual TMX example (English → Spanish)
A minimal bilingual TMX pairs one English <tuv xml:lang="en-US"> with one Spanish <tuv xml:lang="es-ES"> inside a single <tu>. The header declares srclang="en-US", segtype="sentence", and datatype="plaintext".
<?xml version="1.0" encoding="UTF-8"?> <!DOCTYPE tmx SYSTEM "tmx14.dtd"> <tmx version="1.4"> <header creationtool="[Tool name]" creationtoolversion="[Version]" segtype="sentence" o-tmf="[Original TM format]" adminlang="en-US" srclang="en-US" datatype="plaintext"/> <body> <tu tuid="1"> <tuv xml:lang="en-US"><seg>Sign in to continue.</seg></tuv> <tuv xml:lang="es-ES"><seg>Inicia sesión para continuar.</seg></tuv> </tu> </body> </tmx>
Multilingual TMX example with creationdate and changedate
A multilingual TMX adds a third or fourth <tuv> inside the same <tu>, each with its own xml:lang. The creationdate and changedate attributes, formatted as YYYYMMDDThhmmssZ (ISO 8601 basic, UTC), timestamp when each variant was created and last edited.
<tu tuid="42" creationdate="20240115T093000Z" changedate="20240902T144512Z"> <prop type="domain">legal</prop> <tuv xml:lang="en-US"><seg>By signing, you agree to the terms.</seg></tuv> <tuv xml:lang="es-ES"><seg>Al firmar, acepta los términos.</seg></tuv> <tuv xml:lang="fr-FR"><seg>En signant, vous acceptez les conditions.</seg></tuv> <tuv xml:lang="de-DE"><seg>Mit der Unterschrift akzeptieren Sie die Bedingungen.</seg></tuv> </tu>
How it works
How do you use a TMX file in a professional translation workflow?
1
How do you import a TMX file into a CAT tool?
To import a TMX file, open the CAT tool's translation memory manager, create or select the target TM, and run the Import command pointing at the .tmx file. Trados, memoQ, and Phrase all read TMX 1.4 natively from an Import dialog. OmegaT reads TMX by folder placement — the TMX file is dropped into the project's /tm subfolder tree, which means no manual import step is required; the engine picks it up automatically on project load.
- Open the CAT tool's TM manager (Translation Memories view in Trados; Resource Console in memoQ).
- Create a new TM or select the target TM with matching source and target locales.
- Run Import and point at the
.tmxfile. - Choose whether to overwrite, merge, or keep newer segments on duplicate TUIDs.
- Confirm the segment count after import matches the source TMX.
2
How does segtype affect fuzzy matching when you leverage a TMX?
The <header> segtype attribute — block, paragraph, sentence, or phrase — tells a receiving CAT tool how the original was chunked. A sentence-segmented TMX leveraged against paragraph-segmented source yields fewer whole-segment matches and more fuzzy matches below the 75 % floor. The segtype value is therefore one of the most consequential header decisions a localization engineer makes, because it directly controls the quality and quantity of leverage a translator sees on every new project.
| segtype value | Chunk size | Best for | Trade-off |
|---|---|---|---|
block | Whole blocks (paragraphs + headings) | Highly formatted DTP sources | Lowest reuse; rare exact matches |
paragraph | One paragraph per segment | Marketing copy, short articles | Mid-range reuse; context preserved |
sentence | One sentence per segment | Technical docs, UI strings, most production work | Highest reuse; context fragmented |
phrase | Sub-sentence phrases | Software strings, menu items | Maximum reuse; risk of mis-match on polysemy |
3
How do you export a TMX from Trados, memoQ, Phrase, and OmegaT?
Each CAT tool exports TMX from its TM manager. The menu paths differ but the output is TMX 1.4 in every case, because the standard is what every downstream tool expects to receive.
- Trados Studio — Translation Memories view → right-click the TM → Export → choose TMX 1.4.
- memoQ — Resource Console → Translation memories → select → Export → TMX.
- Phrase TMS — Translation memory detail screen → Export → TMX.
- OmegaT — Writes
project_save.tmxplus level-1 and level-2 variants on every save; no manual export step is needed. - Wordfast — TM → Tools → Export to TMX.
- Smartcat — TM detail → Export → TMX.
4
Does TMX leverage bypass the LQA quality gate on certified lanes?
TMX leverage does not bypass LQA. Every finished segment — including 100 % and context matches from a trusted memory — passes through an LQA quality gate priced against an MQM-style scorecard aligned to ISO 17100. Prior errors travel inside a TM just as reliably as prior correct translations, which is why leverage is a productivity input and not a quality guarantee. The LQA pass is where those legacy errors are caught and corrected before delivery.
Which TMX version should you use, and which tools support it?
Use TMX 1.4b. TMX 1.4b achieves universal round-trip across every production-grade CAT tool — Trados, memoQ, Phrase, OmegaT, Wordfast, Smartcat, MateCat — while the never-ratified TMX 2.0 draft was abandoned. 1.4b is the format the industry writes and reads.
| CAT tool | Reads TMX 1.4b | Writes TMX 1.4b | Notes |
|---|---|---|---|
| Trados Studio | Yes | Yes | Desktop market default |
| memoQ | Yes | Yes | Resource Console export |
| Phrase TMS | Yes | Yes | Cloud TMS import/export |
| OmegaT | Yes | Yes | Folder-based placement; levels 1 and 2 on save |
| Wordfast | Yes | Yes | Classic and Pro editions |
| Smartcat | Yes | Yes | Browser-based import |
| MateCat | Yes | Yes | Open-source, browser-based |
| Across | Yes | Yes | Enterprise desktop + server |
Which CAT tools read and write TMX 1.4b natively?
Trados Studio, memoQ, Phrase TMS, OmegaT, Wordfast, Smartcat, MateCat, and Across all read and write TMX 1.4b natively. Interoperability holds because each tool implements the same LISA OSCAR element tree. Vendor extensions stay inside <prop> rather than new elements, so a stranger tool safely ignores them. The full interoperability map across XLIFF, TMX, and TBX lives Computer-Assisted Translation Tools: The Working Translator's Field Guide.
Why did TMX 2.0 never ship?
TMX 2.0 never shipped because LISA dissolved in 2011 before the draft reached ratification. The industry had already converged on TMX 1.4b plus XLIFF 2.0 for the bilingual-interchange side. No production CAT tool writes TMX 2.0. Treat any reference to "TMX 2.0" in a vendor pitch as marketing, not a real artifact.
How does TMX differ from XLIFF and TBX?
TMX, XLIFF, and TBX split 3 jobs in the interoperability stack: TMX carries cumulative translation memory across projects, XLIFF carries one live bilingual file in flight, and TBX carries term-base entries. All three are XML; all three are read by every production-grade CAT tool. A TMX file is a memory that outlives projects; an XLIFF file belongs to one job.
| Format | Granularity | Lifecycle | Payload |
|---|---|---|---|
| TMX 1.4b | Segment pair | Cumulative across projects | Source + target + inline markup |
| XLIFF 1.2 / 2.0 | Segment in file context | One project in flight | Source + target + state + notes + skeleton reference |
| TBX | Term entry | Cumulative, slow-change | Term + definition + POS + usage + forbidden forms |
TMX vs XLIFF: which carries what?
XLIFF carries the live bilingual file for one project — one source-and-target pass between CAT tools or a TMS — while TMX carries the cumulative memory across projects and tools. A CAT tool reads both but writes them for different jobs. The XLIFF side is covered XLIFF: The XML Localization Interchange Format Translators Actually Ship.
TMX vs TBX: translation memory vs term base
TMX stores segment-level source-target pairs; TBX stores term-level entries with definitions, part-of-speech, usage notes, and forbidden forms. A sentence match belongs in TMX. A single-term glossary entry belongs in TBX. Mixing them — stuffing glossary terms into a TMX as one-word TUs — corrupts fuzzy-match scores on receiving tools.
How do you convert other file formats into TMX?
Converting other formats into TMX takes 2 routes: round-trip through a CAT tool that imports the source format and exports TMX, or run a dedicated aligner that pairs source and target documents into aligned <tu> units before export. Both routes produce a TMX 1.4 file — an XML document whose every translation unit contains one <tuv> per language, each identified by an xml:lang attribute such as en-US or de-DE. The choice of route depends on what you are starting from: a CAT-tool round-trip suits bilingual files, while alignment suits existing translated documents where no bilingual source was ever saved.
How do you convert an Excel bilingual list to TMX?
To convert an Excel bilingual list to TMX, place source segments in column A and target segments in column B, label each column header with its BCP 47 locale code (e.g., en-US, es-ES), and save the file as UTF-8 CSV to preserve non-Latin characters. Import the CSV into a CAT tool's Excel-to-TMX converter, map the columns to source and target languages, and export the resulting TM as TMX 1.4. The exported file is a standards-compliant XML document that any receiving tool can import without further transformation.
- Column A header: source locale code (e.g.,
en-US); Column B header: target locale code (e.g.,es-ES). - Rows 2 and beyond: one source segment per row, with its translation in the adjacent column.
- Save as UTF-8 CSV to preserve non-Latin characters such as Arabic, Chinese, or Cyrillic.
- Open the CAT tool's Excel/CSV TM importer and map columns to languages.
- Export the imported TM as TMX 1.4.
How do you align existing translated documents into TMX?
Alignment is the process of pairing a source document with its finished translation sentence by sentence to produce a set of <tu> translation units that can be exported as a TMX file. Dedicated aligners ingest two documents, propose segment pairings automatically, and let a reviewer correct any rows where the automatic pairing drifted. Automated alignment on long paragraphs is prone to one-segment drift, which poisons every affected <tu> in the exported memory — reviewing every proposed pairing before export is therefore essential. The result is a TMX 1.4 file that behaves identically to a memory built inside a CAT tool, and any receiving tool treats it as such. Carriage-return characters inside segment text require special handling: the safest approach is to convert any bare
characters to their XML entity equivalent ( ) before export, which prevents well-formedness errors on import.
What are the best practices for keeping TMX files clean and portable?
4 practices keep a TMX file clean and portable: write TMX 1.4b with UTF-8 encoding, keep segtype consistent across projects contributing to the same memory, strip vendor-specific <prop> elements before handoff, and validate against the TMX 1.4 DTD before sending a memory to a client or new tool. Each practice addresses a distinct failure mode that causes silent data loss or import rejection on the receiving side.
- Write TMX 1.4b — not an unreleased 2.0 — for every export; version 1.4b is the only revision that achieves universal round-trip interoperability across production CAT tools.
- Encode as UTF-8 and declare it in the XML prolog; any other encoding risks silent character corruption on import, particularly for non-Latin scripts such as Chinese, Arabic, or Cyrillic.
- Keep
segtypestable across all projects contributing to the same TM; a memory that alternates betweensentenceandparagraphsegmentation produces unreliable fuzzy-match scores on every tool that queries it. - Use
<prop>elements to attach typed metadata — project name, client, domain, status — to individual<tu>or<tuv>elements, but strip vendor-only prop types before client handoff so the recipient's tool does not choke on unknown type values. - Validate against the TMX 1.4 DTD before every handoff; validation confirms element nesting, required attributes, and well-formedness, catching broken TUs that would otherwise fail import silently.
- Split memories larger than 500 MB into per-domain or per-year TMX files to keep import and export times manageable across all tools.
- Timestamp every
<tu>withcreationdateandchangedateheader attributesYYYYMMDDThhmmssZUTC form so receiving tools can apply recency weighting during leverage. - When a tool encounters unknown elements or attributes in an imported TMX, the safest behavior is to ignore them silently or raise a non-fatal warning rather than rejecting the file — author your own exports with this tolerance in mind and avoid elements outside the TMX 1.4 element set.
Why does UTF-8 encoding matter more than any other header value?
UTF-8 encoding matters most because any other encoding risks silent character corruption on import. Non-Latin scripts — Chinese, Arabic, Cyrillic, Japanese — degrade to mojibake the moment a receiving tool misreads the encoding declaration. Corrupted segments then propagate through fuzzy matching, poisoning every new project that queries the memory. UTF-8 is the correct encoding for all TMX 1.4b files and requires explicit declaration in the XML prolog so there is no ambiguity regardless of which tool reads the file.
How do you validate a TMX file before sending it to a client?
Validate a TMX file against the TMX 1.4 DTD using a TMX-aware editor or a generic XML validator loaded with the DTD. The validator confirms element nesting, required attribute presence, and document well-formedness before a memory handoff. It surfaces broken <tu> records — including unclosed inline tags from the <bpt>/<ept> pairing set — that would otherwise fail import silently on the receiving side, leaving the recipient with a mysteriously smaller memory than expected.
Where can you download free TMX files to test or seed a translation memory?
4 public collections publish free TMX files for testing or seeding a translation memory: the European Commission's DGT Translation Memory covering 22 EU languages, the United Nations Parallel Corpus spanning 6 official UN languages, the OPUS open parallel corpora across hundreds of language pairs, and TAUS Data Marketplace samples tagged by industry domain. All are distributed as TMX 1.4 files under open or permissive licenses, which means they can be imported directly into any CAT tool's translation memory manager without conversion.
- DGT-TM — 22 EU official languages, legislative and regulatory domain, open license; the largest publicly available multilingual TMX corpus and an effective seed for EU-language pairs in legal or institutional contexts.
- UN Parallel Corpus — 6 UN official languages (Arabic, Chinese, English, French, Russian, Spanish), diplomatic register; useful for seeding high-formality language pairs.
- OPUS open corpora — curated parallel text across hundreds of language pairs drawn from subtitles, software documentation, and legislative sources; the breadth of domain coverage makes it valuable for testing how a tool handles large TMX imports.
- TAUS Data Marketplace samples — industry-tagged TMX samples under permissive terms; well-suited for testing domain-specific leverage scenarios before committing to a full data purchase.
When using any public TMX to seed a production memory, validate the file against the TMX 1.4 DTD before import, audit a sample of <tu> records for translation quality, and assign the memory a lower penalty weight than your proprietary TM so its matches do not override in-house validated content. Public corpora are reliable for testing tool behavior and measuring import performance; they are not a substitute for a client-approved, LQA-validated production memory.
How we handle TMX at this site
We ship TMX 1.4b exports for every certified-lane project that leverages a translation memory, and we ship nothing else. TMX 1.4b is the current stable version of the standard — finalized in 2005 — and it is the only revision that round-trips cleanly across every production CAT tool in use today. We do not emit any unreleased draft format, and we do not stuff glossary-sized term entries into a TMX in place of a TBX, because single-word translation units skew fuzzy-match scores on every tool that queries the memory.
Every TMX file we produce is UTF-8 encoded, declares its encoding in the XML prolog, and validates against the TMX 1.4 DTD before it leaves our systems. The <header> block carries all seven required attributes — creationtool, creationtoolversion, segtype, o-tmf, adminlang, srclang, and datatype — plus creationdate and changedate timestamps in UTC so receiving tools can apply recency weighting accurately. We strip vendor-specific <prop> types before any client handoff so the recipient's tool does not encounter values outside the agreed schema.
We refuse to ship a leveraged translation without an LQA pass. A 100 % match from a trusted memory still carries forward prior errors; the LQA scorecard aligned to ISO 17100 is where those errors are caught and corrected before delivery. Translation memory is a productivity tool, not a quality guarantee, and our workflow treats it as such at every stage.
Frequently asked questions
What is a TMX file?
A TMX (Translation Memory eXchange) file is an XML-based file format used to store and share translation memory data between CAT tools and translation management systems. Each TMX file contains translation units (TUs), which pair a source-language segment with one or more target-language equivalents, along with metadata such as language codes, creation date, and tool name. TMX is the industry-standard format for portability — if you switch from SDL Trados to memoQ, for example, you export your TM as a TMX and import it into the new tool without data loss.
What is the TMX file format?
The TMX file format is an open XML standard (originally defined by LISA, now maintained by GALA) designed to exchange translation memory content across different CAT tools and TMS platforms. Files use a.tmx extension and are UTF-8 or UTF-16 encoded, making them human-readable in any text editor. The structure consists of a <header> element (tool, source language, date) and a <body> containing <tu> (translation unit) elements, each holding <tuv> (translation unit variant) segments for every language. TMX 1.4b is the version most tools support today.
What does TMX stand for?
TMX stands for Translation Memory eXchange. The name reflects the format's core purpose: a vendor-neutral, open standard that lets translators and agencies exchange accumulated translation memory data freely between different software environments. It was developed under the OSCAR (Open Standards for Container/Content Allowing Re-use) initiative and has been the de-facto interoperability standard for TM data since the early 2000s.
What is the TMX file format used for?
The TMX format is used to store, share, and reuse translation memory segments across CAT tools, TMS platforms, and machine-translation engines. Practically, translators use TMX files to migrate TM data when switching tools, to share client-specific memories with colleagues or agencies, to train or fine-tune MT engines, and to merge memories from multiple projects. For certified translation workflows, a well-maintained TMX can ensure consistent terminology and phrasing across recurring document types like birth certificates or legal agreements.
What is translation memory?
Translation memory (TM) is a database that stores previously translated source segments paired with their approved target-language equivalents for reuse in future projects. When a new segment is a 100% match or fuzzy match (typically ≥75%) with a stored segment, the CAT tool proposes the existing translation, saving time and improving consistency. For certified translation production, TMs are especially valuable for recurring boilerplate language — legal disclaimers, standard form fields, official titles — where precision and consistency are both required and verifiable.
What is the purpose of using translation memory in CAT tools?
Translation memory in CAT tools speeds up work and enforces consistency by automatically surfacing previously approved translations whenever a matching source segment appears. CAT tools rank matches by similarity: 100% (exact) matches can be inserted instantly, while fuzzy matches (75–99%) give translators a starting point to edit. For agencies producing certified translations at volume — think passport translations, legal contracts, or immigration documents — TM leverage directly reduces per-word costs and minimizes the risk of inconsistent terminology across a client's document set.
How to open TMX files?
TMX files can be opened in any CAT tool (SDL Trados, memoQ, Wordfast, OmegaT), any TMS that supports TM import, or a plain text/XML editor like Notepad++, VS Code, or oXygen XML Editor. To inspect or edit content, right-click the file and choose 'Open with' a text editor — the XML structure is human-readable. To actually use the TM for translation, import the.tmx file into your CAT tool's TM database via its import wizard, which takes under a minute for files up to a few hundred MB.
How do I edit a TMX file?
You can edit a TMX file directly in an XML or text editor (Notepad++, VS Code, oXygen) by modifying the <seg> content inside each <tuv> element, or by using a dedicated TMX editor such as Heartsome TMX Editor (free) or the built-in TM editors in Trados or memoQ. For bulk corrections — fixing a systematic terminology error across thousands of TUs — a find-and-replace in a text editor on the raw XML is the fastest approach. Always validate the XML structure after manual edits to avoid import errors in your CAT tool.
How to make a TMX file?
You create a TMX file by exporting translation memory from a CAT tool or TMS — virtually every major tool (Trados, memoQ, Wordfast, OmegaT) has an 'Export TM as TMX' function. Alternatively, you can build one programmatically: since TMX is plain XML, you can write or generate the <header> and <tu>/<tuv> structure with any scripting language and save the file with a.tmx extension. For agencies starting from bilingual Excel glossaries, converting spreadsheet data to TMX via a script or a tool like Okapi Olifant is a common bootstrapping approach.
How to convert an Excel file to TMX?
To convert an Excel bilingual table to TMX, use a free tool like Okapi Olifant or the Rainbow utility from the Okapi Framework, which accept tab-delimited or Excel input and export valid TMX. The process involves: (1) structuring your spreadsheet with one column per language, (2) exporting as CSV or TSV, (3) importing into Olifant and mapping columns to source/target language codes, then (4) exporting as TMX 1.4b. Some TMS platforms also offer direct Excel-to-TM import wizards. The resulting TMX can then be loaded into any CAT tool.
Related
- Translation Quality: What It Means and How Production Teams Measure It
- Desktop Publishing Services for Translators: Scope, Tools, and Rates
- XLIFF: The XML Localization Interchange Format Translators Actually Ship
- Translation Management System
- ATA Membership: Tiers, Dues, and What Translators Actually Get
- Back Translation: How It Works, When to Use It, and Its Limits