Generalizations.HomogeneityProjection — cross-paper data pool #

Cross-paper test substrate for how the homogeneity gap of plural definites projects under embedding quantifiers and operators. The pattern's empirical literature spans [Lob00], [Gaj05], [Spe13a], [Mag14], [KC15], [Kri16], [BL21], [ABB+23], [KS21b], and [Has25a]; the empirical generalisation predates any one formal account, justifying a theory-neutral Data/Generalizations/ anchor.

Main declarations #

EmbeddingOperator — operator labels recurring across the literature.
GapScenario — scenario classification used by the C-series experiments (TRUE / FALSE / GAP / GAP? / GAP??).
ProjectionDatum — typed empirical datum; lifted from the raw LinguisticExample rows by fromExample.
Examples (generator-managed) — pooled stimulus rows from each paper that contributes projection-relevant data; per-paper JSON inputs live in Linglib/Data/Examples/<Paper>.json, and this hub imports each paper's generated Data.Examples.<Paper> module (see scripts/gen_examples.py).
allData — the test pool: every LinguisticExample whose paperFeatures carry the keys this hub recognises.

Implementation notes #

This file is the data side of cross-account testing of homogeneity projection: a ProjectionDatum per observed outcome, pooled in allData. Accounts state their predictions in their own study files, at the granularity their paper supports, and are run against the pool there — restricted by source.bibkey to papers available at each study's publication date. Decidable per-datum and divergence theorems live in those study files too.

The substrate is restricted to the smallest set of operator cases that closes the current ≥2-consumer graduation criterion: every / no (every major contribution tests these), exactlyTwo (introduced by [KC15] Exps. A3/B3/C3/C4), and notEvery (the no/notEvery asymmetry from [ABB+23]). Extend as new consumers land.

There is deliberately no projector-stipulated monotonicity : EmbeddingOperator → ContextPolarity function — the classical mapping (notEvery ↦ downward by De Morgan) disagrees with the empirical observation that notEvery patterns with every rather than no in QUD-manipulation acceptance ([ABB+23]), so any single mapping would be wrong for at least one consumer.

Todo #

Run the rival accounts postdating [KC15] against this pool: the exhaustification account ([BL21]) and the mature supervaluation/trivalence accounts ([Kri16], [KS21b]). The account variants the paper itself assesses — [Mag14]-style implicature construals, [Spe13a] supervaluation, and [Sch94] / [Lob00] / [Gaj05] presupposition with universal projection — are already run against it in Studies/KrizChemla2015.lean, restricted to the paper's tested cells.
Pool a second paper's rows (JSON-ify the [ABB+23] acceptance data) to close the ≥ 2-papers admission prong.
Add a denotation hook EmbeddingOperator → ∀ α, Quantification.GQ α once accounts derive predictions structurally rather than dispatch on label (per [PW06] discipline).

Generalizations.HomogeneityProjection — cross-paper data pool #

Main declarations #

Implementation notes #

Todo #

Substrate #

Datum schema #

LinguisticExample adapter #

Pool #

`LinguisticExample` adapter #