XLIFF: Moving Content Through a Localization Workflow

What makes an interchange useful

A translator needs more than visible words. The receiving system may also need to retain identifiers, formatting boundaries, context and production information. The practical goal is a translation that can return to the product without losing its connection to the original content.

This differs from storing previous translations in a memory or maintaining approved terminology. Those jobs are discussed in TMX and TBX.

An illustrative round trip

A website exports a password-reset message containing a link and a user-name variable. Our recommended test is to translate a small representative sample, import it back and inspect the rendered page.

Check that the message is attached to the correct interface element, the variable remains usable, the link still works and the translation fits its context. Then test the actual file version and features that the sending and receiving tools claim to support.

This is an original interoperability exercise. It is not an official XLIFF conformance test, and merely passing it does not establish complete support.

Version and ISO-reference caution

There is a discrepancy in the checked public records. OASIS describes ISO 21720:2024 as its ISO version of XLIFF 2.1, while ISO's public abstract says version 2.0. We do not resolve that discrepancy by guessing. The ISO edition is published, but the exact mapping must be verified against the applicable full text before an implementation or contractual claim. ISO 21720:2024.

Relationship to LISA and AI

XLIFF belongs to the same interoperability landscape as LISA's resource formats, but it is an OASIS development, not a LISA-authored specification.

For AI-assisted translation, our editorial advice is to keep structural content under controlled parsing and validation. A model-generated target can change placeholders or markup even when it reads naturally. Validate structure independently and inspect meaning separately.