Source
Dzhafarov, Zhang & Kujala (2015), LIT-264, §§1–8. Supporting: Budroni et al. (LIT-263, §IV.C.5) on the same criterion in physics.
What was actually shown
Contextuality-by-Default gives each object a separate random variable in each context and calls a system noncontextual when some coupling makes those variables coincide as often as each pair’s own marginals allow (§1, Eq. 7). With consistent marginals its cyclic criterion reduces to the Leggett–Garg, CHSH and KCBS inequalities (§2). Applied to 73 poll question-order pairs (§3), Schröder-staircase judgements (§4), the Aerts–Gabora–Sozzo animal/sound choices (§5), 23 primed ambiguous word combinations from Bruza et al. (§6) and psychophysical matching (§7), the criterion finds no contextuality, apart from one poll pair the authors discount for multiple comparisons. The reported CHSH violation in §5 assumed consistent marginals; once inconsistent marginals are accounted for, it disappears. The Hilbert-space model of question-order effects predicts noncontextuality (§3). In physics the same criterion still finds KCBS data contextual (LIT-263 §IV.C.5). “The same criterion” holds for binary cyclic systems. The review presents Kujala et al.’s maximally noncontextual models as one noise model among four, does not name Contextuality-by-Default, and says nothing about behavioural data (NOTE-844).
What this does not say
- That behavioural systems are never contextual; that is the authors’ working hypothesis, and later Contextuality-by-Default work is reported to find behavioural contextuality (not held, not read).
- That quantum-like models of cognition or retrieval (van Rijsbergen, LIT-262) are wrong; the paper says such a model predicts the absence of contextuality.
- Anything about Kochen–Specker or Spekkens contextuality in physics.
Connections
- The two tests of contextuality in language data the record has read (LIT-842, LIT-844) are filed as THEORY-623. Corpus meaning-selection models are all signalling, their two Contextuality-by-Default verdicts are not secured by the paper’s own bootstrap, and the BERT anaphora models have their support fixed by the schema. Neither is a robust counterexample, and neither uses human judgements.