The problem behind the design
A revised document contains both changed and unchanged content. A production team needs a reliable way to understand what happened between versions and what previous work remains relevant.
Our practical interest is the distinction between text similarity and document identity. Two identical sentences can occur in different places with different roles. Conversely, a sentence in the same location may have changed enough to require a new translation decision.
An illustrative revision
A manual contains the same warning in two chapters. In the next release, only the second instance changes. A team comparing plain strings may overlook why that specific occurrence matters.
Our suggested exercise is to preserve identifiers and compare successive source snapshots. Record which units changed, which remained stable and which were moved or deleted. Then inspect how that evidence affects translation reuse.
This scenario explains the underlying need. It is not an implementation of xml:tm or a claim that generic identifiers satisfy its requirements.
Relationship to other formats
xml:tm should not be used as a synonym for TMX. It addresses document-linked memory, whereas TMX is an interchange reference for translation-memory data. Its historical specification discusses relationships with segmentation, translation rules and localization interchange, but those relationships must be read in the versions and context of that document.
Historical status
xml:tm belongs directly to LISA's preserved portfolio. ETSI's later work register records the XML Text Memory project as stopped. That history does not establish present-day adoption or a maintained implementation path. ETSI LIS work programme.
For a new system, our recommendation is to start with actual format support and lifecycle needs rather than choosing a historical reference on name recognition alone.
AI and change tracking
In an AI-assisted workflow, stable source snapshots and identifiers can help show which inputs produced which outputs. Regenerating an entire document without recording changed inputs can make review harder. This is a general design lesson, not a claim that xml:tm standardizes model versioning or AI provenance.
Source record and related reading
xml:tm 1.0 and the ETSI work record above. The historical specification is available under its stated CC BY 3.0 terms. The example and recommendations here are original commentary.