Topics — What a work is about. Nucleation holds whatever reading the owner files here, whether or not an Anthology of the SOTA topic could also hold it; instruction for ML practice belongs to the anthology (ADR-048). The words are seeded from the clusters the reading-time triage of 2026-09-25 found, and grow by decision.
Linguistics — the study of language itself — morphology, lexicalism, word segmentation, syntax and semantics as linguists pose them, not as models encode them.
8 of 371 QUESTION documents. Back to the full index.
| # | Title | Summary | Status |
|---|---|---|---|
| QUESTION-002 | What survives when an utterance is translated, retold or rendered in another medium? | The manuscript’s title question. An answer has to say what is held fixed across the transformation — wording, proposition, or something relational — and how preservation can be measured rather than asserted. | Open |
| QUESTION-014 | Does a stable speaker footing measurably reduce pragmatic drift relative to shifting frames, even when both drift lexically? | A67, after the inference harness could not be run: “The next result worth obtaining”. The manuscript’s Case II (stable complicit, stable admonitory, alternating and random frames) would answer it; no data exists. | Open |
| QUESTION-021 | Is there a frame-independent communicative act that translations approximate, or is the act’s identity constituted by a family of frame-relative realizations? | Opened at A18 as “the deepest question in your line of thought”. The manuscript takes the second horn in practice (CLAIM-115) but never argues against the first. | Open |
| QUESTION-023 | Can the opacity of insider vocabulary be a preservation target in its own right? | Raised twice by the assistant in the Henley workshop (A6, A9) and never answered. The manuscript treats opacity only as a loss to an audience lacking knowledge (§5), not as something a rendering might owe its reader. | Open |
| QUESTION-137 | Do human transmission chains converge to the learners’ independently measured prior where their responses behave as posterior samples, and depart from it as the MAP analysis predicts where they behave as maximizers? | The empirical half of what THEORY-155’s promotion condition asked for until 2026-10-19. The theorem, that an ergodic chain of posterior- sampling Bayesian learners with a shared prior and matched production converges to that prior, is proved and Active. Whether human chains behave so is a separate question, and only a chain whose learners’ prior is measured or set apart from the chain can answer it. | Open |
| QUESTION-209 | Do listeners in reference and signalling games invert an informative speaker when alternatives and costs are manipulated with the uttered form held fixed, the model’s parameters are set in advance, and its predictions are tested on trial-level data against a literal listener and a likelihood-only model? | The promotion condition formerly on THEORY-172. THEORY-172 was promoted on 2026-10-20 (ADR-047 v3) on what was measured: in Frank and Goodman’s reference games listeners’ mean bets matched a parameter-free inverted informative speaker with a measured salience prior, and in Bergen, Goodman and Levy’s signalling games listeners read a novel signal by its alternative. The evidence is a fit of condition means with no rival model fitted. | Open |
| QUESTION-214 | Do colour-naming systems stay near the information-bottleneck bound when the need distribution is measured independently of the naming data, and do lexicons in a second semantic domain? | The promotion condition formerly on THEORY-163. THEORY-163 was narrowed to what Zaslavsky, Kemp, Regier and Tishby measured and promoted on 2026-10-20 (ADR-047 v3): the World Color Survey languages and English lie near the information-bottleneck curve at a trade-off parameter just above 1, nearer than hue-rotated variants of themselves, with a need distribution estimated from the naming data. A test with an independently measured source, or in a second domain, is open. | Open |
| QUESTION-233 | Do language models show the data-limited exponent γ/(2β) because data limits the usable horizon: does the exponent track γ/(2β) when one corpus statistic is moved and the other held, and does the match survive at horizons of thousands of tokens with γ estimated outside the model class? | Filed when the promotion audit (batch 062, ADR-047 v2 and v3) narrowed and promoted THEORY-204. That the fitted data-limited exponent matches γ/(2β) on TinyStories and WikiText-103, and that assumed per-lag scaling forms give min(δ, γ/(2β)), is settled. Whether real learners show the exponent for the horizon reason was the old promote_when, and it moves here. | Open |