nucleation

CLAIM-068: Repeated reconstruction is attraction toward conventional regions of communicative possibility: pragmatic stances, not only lexicon and grammar, can be stabilized by transmission

4 min read

Properties13
StatusProposed
Version3
Rolethesis
Warrantinterpretive
Tagsphilosophy-of-language, cognition
Linepragmatic-transport
Rests onCLAIM-090 — Serial transmission is reconstructive and filtered by the transmitters' inductive biases, so a stable endpoint can be far from the ancestral utterance while remaining interpretable
Corroborated byTHEORY-173 — Large models are a cultural and social technology rather than agents: lossy, uninvertible summaries of human-produced information that, like prices and bureaucratic categories, let it be reorganized at scale, THEORY-164 — As Raviv, Meyer and Lev-Ari report it, in a laboratory game where 12 fully connected groups of eight and 12 of four built a new communication system with total interaction held fixed, the larger groups' languages grew more systematic faster, ended more systematic and differed less from one another, while communicative success and convergence did not differ by size; input variability, higher in the larger groups, predicted the next round's gain in structure, an association measured and not manipulated
UsesTERM-027 — cultural attractor
Illustrated byCASE-068 — Risk messages passed down diffusion chains: the evaluative signal kept and amplified while the content decays (Moussaïd, Brighton and Gaissmaier; Jagiello and Hills)
Cited as “concludes” byARG-027 — The record's iterated-learning readings show that pragmatic stances and footing are stabilized by transmission
Cited as “rests on” byCLAIM-059 — Repeated reconstruction chains fall into different regimes under conditions assigned in advance, contracting toward conventions, accumulating neutrally or amplifying, and amplification needs a metric other than total variation, an enlarged state or state-dependent dynamics
Rendered intodocs/arguments-by-conclusion.md

Graph View

The claim

A50 §3: “Repeated retellings may converge toward culturally familiar narratives, conventional speech acts, or highly stereotyped interpretations … translation drift isn’t random wandering. It’s an attraction toward certain regions of communicative state space.” The running case: “the criminal-world complicity might survive several stages while the historical references disappear rapidly. Or the poem might gradually transform from a bawdy celebration of predictable excess into a moralizing warning about financial irresponsibility”, which is “communicative drift along a stance dimension, not merely lexical degradation.” A50’s questions: convergence to a stationary distribution; multiple attractors for genres and stances; how fast information about the original decays; which properties last longest.

A52 §7.4 named the selection pressures (memorability, efficiency, genre conformity, social acceptability, learnability), and §7.5 the hypothesis “that cultural transmission can stabilize characteristic pragmatic stances”, extending “iterated learning from lexical and grammatical structure to speaker footing and communicative conventions”. A52 §9.4: “Distinguish fidelity to a particular original from convergence toward broadly useful communicative conventions.”

Where it went

The manuscript §8 keeps “A stable conventional endpoint may be far removed from the ancestral utterance while remaining interpretable” (CLAIM-090) and the stance example (“from speaker–audience solidarity to authoritative advice”). It does not keep attraction, the selection pressures or the footing hypothesis. The record’s readings qualify the iterated-learning premise (CLAIM-066).

C1 Appendix D, the first full draft (U22), qualified it: “The long-run structure of observed utterances is therefore a property of the specified transmission mechanism, not a generic consequence of repetition”, depending on whether the chain has a unique stationary law or several absorbing classes. E1’s contractive chain (CASE-017) is the stipulated instance of convergence to a convention (CLAIM-081).

Proposal v5 (A82 §8.5–8.6, under U30) added two points. The first: “Distinguish convergence of realizations from convergence of observable probability distributions.” The second: retellings of a criminal ballad “may progressively eliminate observables associated with its original subculture”, while modern adaptation “can introduce new distinctions”, so the cover itself drifts. C6 kept “cultural attractors” as a section heading (§8). The manuscript keeps only “change the available measurement cover” (§11).

The crystallized argument (A110 §20) gave the attractor its reading: “a shared decoding convention rather than a preserved historical object” (TERM-027). The manuscript never uses the word.

Three readings since

  • Two attractors. Kalish et al. (LIT-820) find positive and negative linear functions both act as endpoints.
  • Group size. Raviv, Meyer and Lev-Ari (LIT-825, THEORY-164) find that larger groups converge on more systematic and more similar languages, while small groups are more open to drift. This is horizontal interaction, not a chain.
  • Structure in the corpus. Propp (LIT-828, THEORY-167) argues that a teller is bound in the order of functions but free in characters, attributes and wording. That is a structural thesis about the corpus, not a measurement of retellings.

A mechanism when a model is in the loop

Farrell, Gopnik, Shalizi and Evans (LIT-848, THEORY-173) give a mechanism for contraction when a large model takes part in transmission. A model fitted to reproduce text well on average is least accurate on rare material, so it “might” homogenize culture. The paper states this but does not test it, so it is weak support.

Note of 2026-10-10: the replies

CLAIM-234 is conceded: no reading the record holds measures a stance or footing along a chain, so the stance extension has no study under it. The defeat condition is unchanged. Which transmission-chain studies of social and affective content to read is left to the owner.

Note of 2026-10-12: edge for a prose citation

THEORY-173 was already cited in this claim’s body as weak support. It is now a typed ground.

Note of 2026-10-12: the scope of a ground added on 2026-10-12

THEORY-173 became a typed ground of this claim on 2026-10-12, as the note above says. It also grounds CLAIM-234, the granted objection to this claim, which calls it “a mechanism stated and not tested”. The mechanism it gives is contraction: homogenization when a large model takes part in transmission. It bears on this claim’s attraction toward conventional regions, not on the extension to footing, which CLAIM-234 says “has no study under it”.

Note of 2026-10-29: what holds it back

Its premise CLAIM-090 is Proposed: no chain in the record ends on a prior measured on its own transmitters before transmission (QUESTION-137). Its defeat condition asks for chains started from different stances, and no such study is held. ARG-027, which argued that the iterated-learning readings show stance stabilization, is Rejected. CASE-068’s diffusion chains show an evaluative signal kept and amplified, not chains from different stances converging.

Backlinks