The objection
CLAIM-111 (A45 §6): P_o(z | c_o, u_o) ≈ P_t(τ_z(z) | c_t, u_t), “where τ_z maps corresponding latent communicative frames across linguistic environments”. Its defeat condition tests renderings “whose inferred-situation posteriors match those of the source”. Its worked model is the rational-speech-act listener of THEORY-172, “a posterior over what the speaker meant, given the utterance and its alternatives” (CLAIM-111).
THEORY-172’s title says what that costs: “the interpretation of a fixed form depends on what else the speaker could have said”. Its model: P_L(m|u) ∝ P_S(u|m)P(m), where P_S chooses among the utterances true of m in proportion to exp(α[log P_Lit(m|u) − c(u)]), and P_Lit conditions the prior on literal truth. The normalization of P_S runs over the set of alternative utterances. A translation does not keep that set: it moves the utterance into another language, whose alternatives are different.
A worked case. Two meanings, m₁ and m₂, with a uniform prior; α = 1 and no costs.
- Source language: two utterances. a is true of m₁ and m₂; b is true of m₂ only. Then P_Lit(m₁|a) = P_Lit(m₂|a) = 1/2 and P_Lit(m₂|b) = 1. The speaker: P_S(a|m₁) = 1, since a is the only utterance true of m₁; and P_S(a|m₂) = (1/2)/(1/2 + 1) = 1/3, P_S(b|m₂) = 2/3. The listener: P_L(m₁|a) = (1 · 1/2)/(1 · 1/2 + 1/3 · 1/2) = 3/4.
- Target language: one utterance, a′, true of m₁ and m₂, with no counterpart of b. Then P_S(a′|m₁) = P_S(a′|m₂) = 1, and P_L(m₁|a′) = 1/2.
a′ has exactly the literal content of a, and the posteriors differ, 3/4 against 1/2. This is the scalar case (“some” read as “not all” because “all” was available). Whether a rendering matches the source’s posterior depends on the alternatives each language offers, so CLAIM-111’s τ_z, which maps frames, is not enough: the criterion also needs a correspondence between source and target alternative sets, or a rule for when a target form with different alternatives counts as matching. That is one more correspondence that has to be fixed from outside both languages, of the kind QUESTION-005 asks about. A search of the claims for “alternative set” and “could have said” on 2026-10-10 finds no entry that names it.
The evidence. THEORY-172 is “Shown for artificial games with given
alternatives”. Its promote_when says what exists cannot settle it: “A
high correlation of condition means with a parameter-free model, which is
what exists, cannot settle it.” So it is a worked model of CLAIM-111’s
criterion, not evidence that the criterion applies to poems, where the
alternatives are neither given nor few.
What it does not say
It does not say the rational-speech-act model is wrong, nor that CLAIM-111 is false. It says that, on the model CLAIM-111 adopts, matching posteriors across languages needs a correspondence of alternatives that the record has not stated. It is minor for the argument: CLAIM-111 is not a premise of CLAIM-115 or CLAIM-139.