nucleation

CLAIM-049: An adaptive interpreter's state includes its message, context knowledge and learned parameters, and learning can make some relational invariants more stable and others less

2 min read

Properties8
StatusActive
Version2
Rolethesis
Warrantinterpretive
Tagsphilosophy-of-language, representation-learning
Linepragmatic-transport
Rests onCLAIM-074 — Fidelity is relative to the interpreter's learning history: what the current context supplies and what earlier experience has left in the interpreter are separate histories, told apart by what they carry and not by whether the change sits in activations or in weights, so fidelity under a fixed interpreter, across interpreters and across states of adaptation are different quantities
Corroborated byTHEORY-155 — Iterated learning by Bayesian agents who sample from the posterior converges, in an ergodic chain, to the shared prior, THEORY-489 — Chains of human function learners, each trained on the previous learner's responses, forget their starting function and in 28 of 32 families converge within a few generations to a positive linear function, the bias earlier studies report, whatever function trained the first learner; this is convergence to the learners' favoured function, not shown to be convergence to their prior

Graph View

The claim

A110 §22: “We can represent the effective state of an interpreter as W_i = (U_i, K_i, θ_i), where U_i is the current message, K_i is contextual knowledge, and θ_i denotes the relevant learned or adaptive parameters. A reconstruction kernel acts on this enlarged state. This matters because the same source expression need not produce the same target distribution under different interpreters. A translator familiar with a historical underworld genre might conserve its affiliative humor. An interpreter without that knowledge might reconstruct it as a generic moral statement about irresponsible spending. Learning can make some relational invariants more stable and others less stable.”

Where it went

Manuscript §11: “an adapted receiver may recover an implicit relation without additional message bits”, and §5’s signal/interpreter distinction (CLAIM-118). The enlarged state, and the claim that adaptation changes which invariants are conserved, are not in it. It is the place where CLAIM-074 meets QUESTION-006, and where CLAIM-267’s amplifying regime becomes possible.

Parallels

In iterated learning the learners’ own bias, not the source, fixes what a chain conserves: Bayesian samplers converge to their prior (THEORY-155), and human function learners to a positive linear function whatever trained the first (THEORY-489). Neither varies the learners’ adaptation, so they show that the interpreter’s state sets what survives a chain, not that learning moves it.

Backlinks