The inference
§8, quoted in CLAIM-090: “Bartlett’s serial reproduction experiments and Bayesian iterated-learning models both show that transmission is reconstructive and filtered by inductive biases rather than neutral copying.” The record reads Griffiths and Kalish in THEORY-155 (NOTE-599) and Kirby, Cornish and Smith in LIT-771 (NOTE-596, Skimmed). Bartlett is unread (LIT-768) and is no premise here.
Critical questions
- Does the work say this, read in context? In part. Griffiths and Kalish prove convergence to the shared prior for learners who sample from the posterior (THEORY-155). Kirby et al. measured no prior and use no Bayesian learners. CLAIM-066 (2026-10-09): “Without the homonym filter (Experiment 1) the languages collapsed to two to five words that underspecified meaning; compositional structure appeared only when the experimenter removed homonyms before each generation (Experiment 2), which is a selection on the data, the case Griffiths and Kalish’s analysis explicitly excludes.”
- Is it said within the scope the claim needs? Only for sampling learners. “Learners who take the most probable language depend also on how much data passes between generations, and can amplify weak biases” (CLAIM-066, from THEORY-155’s MAP case).
- Does the claim lean on the reading for more than the reading holds? Yes. This is the question that fails. CLAIM-066: “The manuscript cites them as showing one thing; they show different things.” Kirby et al. show structure emerging under transmission with an experimenter’s filter, not filtering by measured inductive biases. Which learner model applies decides “whether the endpoint reflects the learners or the transmission filter”.
Proposed, not Rejected. CLAIM-066 “does not undercut the manuscript’s §8 conclusion that a stable endpoint can be far from the ancestor: for sampling learners the stationary distribution is the prior, independent of the starting language.” Whether human chains reach the learners’ independently measured prior is QUESTION-137. THEORY-155 already holds every point CLAIM-066 made about the two sources, including the homonym filter under “It does not cover selection” and the MAP case, so the reading gains nothing.
A chain without a filter
Kalish, Griffiths and Lewandowsky (LIT-820, THEORY-489) supply the unfiltered chain with a bias stated in advance that Kirby et al. lack, and CLAIM-090 already grounds on them. They still cannot tell sampling learners from MAP learners. For evaluative stance, CASE-068’s chain studies bear on CLAIM-090’s measured-prior condition, with a prior reported after the experiment and measured indirectly.