The claim
This is the observation CLAIM-219’s dilemma starts from, stated on its own so that its warrant can be.
Dzhafarov, Zhang and Kujala (LIT-264, read in NOTE-235) apply the Contextuality-by-Default criterion to five kinds of behavioural data, two of them previously published as contextual. “None is contextual.” NOTE-235 separates a response distribution that depends on context, inconsistent connectedness, “a trivial sense of contextuality”, from contextuality proper: “Behavioural data have plenty of the first and, so far, none of the second.”
CLAIM-125, from Wang, Sadrzadeh, Abramsky and Cervantes (LIT-851): “69 of the 90 systems signal” and “34 signal enough that they cannot be contextual at all”. The paper tests none of the 90 for contextuality (LIT-851’s status note).
CASE-053 collects the documented tests of contextuality proper on elicited human judgements. No study of pragmatic or meaning judgements finds it at an adequate sample size: LIT-264 and Wang’s thesis (2024) find none, and Lo, Sadrzadeh and Mansfield’s Winograd schemas, the one test with pragmatic content, find a violation of about one standard deviation, with no-signalling forced by the response format.
What it does not say
- That a contextual pragmatic data set cannot exist. LIT-264 offers the absence of contextuality as a working hypothesis, and CASE-053 documents that the adequate test has not been run.
- That no behavioural data anywhere are contextual. CASE-053 records Cervantes and Dzhafarov’s “Snow Queen” as contextual with high reliability; its content is instructed attribute matching, not pragmatics, and the result is disputed.
- Anything about CLAIM-105’s headline. What follows for it is CLAIM-219’s dilemma and CLAIM-201’s narrowing, which rest on this claim.
- That a contextual result would settle the matter by itself. How much contextuality a data set shows depends on how its outcomes are coded (CLAIM-245), so the coding has to be fixed in advance.