The claim
A18: “Lorentz transformations are invertible, compose according to a precise rule, and preserve a specified mathematical structure. Real translation operations generally have none of those guarantees. They can be many-to-one, culturally asymmetric, historically contingent, and fundamentally ambiguous. So I wouldn’t claim we’ve identified a relativity theory of translation.”
And the second disanalogy: in physics the invariants “arise from the mathematical structure of the theory and empirical constraints. In translation, by contrast, we must discover—or negotiate—which communicative properties deserve invariance.” The manuscript keeps this as the normative choice of observables (§11).
Restated twice. A38 §7.1: “linguistic frames do not possess the canonical transformations or invariant metric of relativistic physics”, and “relativity provides a structural analogy, not physical evidence about linguistic meaning”. A40 §9.2: “Pragmatic frames generally lack canonical mappings”, and “Translation is commonly many-to-many and context-dependent rather than invertible”, with the remedy (§9.3): “Move away from the claim that each translation is simply a coordinate description of one fixed communicative object.” A40 adds the passive/active point (CLAIM-004) as a further undercut.
C1 §2, the first full draft (U22), gave the analogy its last form: “The physical model supplies the methodological question—what relations survive a transformation?—not a literal Lorentz geometry for discourse.”
Proposal v5 planned to “Recover the initial physical analogy” in a closing section (A84 §14.3; CLAIM-004). C6 dropped it, and the manuscript’s §4 replaces relativity with symmetry and equivariance (TERM-004).
What it changes
It leaves ARG-004 its research-programme respect only. It is the premise of CLAIM-098.
Note of 2026-10-28: split under ADR-061
“The relativity analogy holds only as a research programme” is a verdict on ARG-004, and ARG-004’s body and status already state it. Under ADR-061 the claim keeps what it says about translation operations: that they are generally not invertible, do not compose by a fixed rule and preserve no structure given in advance, and that the invariants must be discovered or negotiated. That is the subject-matter claim the decision names as the model of an undercut. The quoted text above is unchanged.