What Is TMX (Translation Memory eXchange)?

TMX (Translation Memory eXchange) is an open XML standard for storing and transferring bilingual translation memory data between localization tools and platforms. A TMX file contains aligned source-target segment pairs — each sentence or phrase you’ve translated, stored alongside metadata about who translated it and when. It exists to make translation memory portable across different software systems.

What TMX Contains

A TMX file is structured XML. The core unit is a <tu> (Translation Unit), which holds one aligned pair: the source segment and its approved target translation. For a Translation Memory with English source and French target, each <tu> contains two <tuv> (Translation Unit Variant) elements — one in English, one in French.

Alongside the text, each Translation Unit can carry metadata:

• The source and target language codes (e.g., en-US, fr-FR)
• The date the segment was created or last modified
• The creation tool or system that generated it
• The user or process that approved it

This metadata is what makes TMX more than a bilingual spreadsheet. A Translation Memory system that ingests TMX can read not just what was translated, but when, by whom, and through which process — information that affects how that segment is weighted when it’s retrieved for future projects.

Why TMX Matters: Portability

The most important function of TMX is vendor independence. Translation Memory is an asset that accumulates value over years of localization work. A large TM — hundreds of thousands of approved segment pairs — represents a significant reduction in future translation costs through exact match and fuzzy match reuse.

Before open standards like TMX, TM data was often stored in proprietary formats — Trados .sdltm, for example — that were difficult or impossible to migrate to another tool. A team wanting to switch platforms risked losing years of accumulated TM data, which in practice created strong vendor lock-in.

TMX is recognized by every major localization tool — CAT tools, TMS platforms, and modern localization APIs. A .tmx file exported from one platform can be imported into another without data loss. This portability is the standard’s primary value: the TM belongs to the organization, not to the tool that built it.

Metadata and TM Quality

When a TM system processes an incoming TMX file, it can use the segment metadata to make decisions about reliability.

A segment generated by a generic machine translation engine and never reviewed by a human carries different reliability than a segment that was translated by a professional linguist and approved through an LQA workflow. Both can exist in the same TMX file — the metadata tells the system which is which.

This matters for how segments get surfaced. A Semantic Reranker — which retrieves TM matches by meaning similarity rather than character matching — can weight human-approved segments higher than unreviewed machine output when surfacing suggestions. A TM that has been built carefully, with consistent approval workflows, becomes a more reliable asset over time.

The practical implication: TMX quality depends on TMX hygiene. If unreviewed machine output accumulates in the TM without being marked as such, the metadata distinction is lost. Segments that should carry a lower confidence score are treated the same as carefully reviewed work. Most localization teams with mature TM practices maintain a clear workflow distinction between provisional machine output and human-approved translations.

TMX vs. XLIFF

Both TMX and XLIFF are XML standards used in localization, but they serve different purposes.

TMX is the archive. XLIFF is the active job. When a localization workflow starts, XLIFF carries the new content to be translated. When that job completes, the approved segments can be written back to the TM and exported as TMX for the long-term record.

Related Terms

• Translation Memory — the database that TMX stores and transfers; the underlying asset
• XLIFF — the complementary file standard for active translation workflows
• CAT Tool — the editing environment that reads and writes TMX files
• TMS — the platform that manages TM assets, including TMX import/export
• Exact Match — the 100% match condition that TMX data enables
• Fuzzy Match — partial matches surfaced from TM data stored in TMX format
• Glossary Management — the term-level complement to TMX’s segment-level data
• Language Service Provider — agencies that maintain and transfer client TM data in TMX format

Related Guides

• Translation Memory: How It Works and Why It Matters — how TM is built, maintained, and how TMX fits into the longer-term asset strategy
• How Flixu’s Translation Memory Works — how Flixu handles TMX import and semantic retrieval
• For Translation Agencies — how agencies manage client TM assets across platforms using TMX

Last Updated: March 2026 · Author: Deniz, Founder — Flixu AI