TMX

TMX is the LISA OSCAR standard for translation memories.

Standard conformance

Summary: TMX version 1.4 conformance to Level 1, except that no markup is stripped.

  • All required header fields are supplied.

  • The adminlang field in the header is always English.

  • None of the optional header fields are supplied.

  • Multiple language variants are supported. Source variants are selected using the configured source language, a translation unit’s srclang, or the header’s srclang, in that order. Target variants can be selected by configuring a target language.

  • No special consideration for segmentation.

  • Currently text is treated as plain text, in other words no markup like HTML inside messages are stripped or interpreted as it should be for complete Level 1 conformance.

Language matching

Language lookup first matches the requested code exactly, ignoring case and _ versus - separators. If a bare language code has no exact match, its Weblate language-data default-country variant is used when present. For example, en can select en-US and ko can select ko-KR. An explicit en entry always takes precedence over en-US.

Regional and script-qualified requests remain exact-only: en-GB does not select en-US, and en-US does not select en. Unmatched languages remain empty; target fallback does not reuse the source node for a distinct requested language. Editing a matched language preserves its existing XML language tag.