Embedded questions like in (1) are the central object of study in the formal semantic analysis of questions (Karttunen 1977, Groenendijk & Stokhof 1982, 1984, Heim 1994, Klinedinst and Rothschild 2011, i.a.). Among others, the readings in (2) are discussed.
(1) Ali knows who danced at the party.
(2) a. Strong exhaustive reading (SE)
Ali knows for all people who danced a the party that they danced and he knows that they are the only ones who danced.
(2) b. Intermediate exhaustive reading (IE)
Ali knows for all people who danced at the party that they danced and he does not have false beliefs about any non-dancers.
For a long time, the dominant position in the theoretical literature was that questions embedded under the verb to know can receive only a SE reading. In experiments, however, the IE readings was dominant (Cremers & Chemla 2016, Cremers et al. 2017).
In my talk, I present an experiment that addresses this discrepancy by investigating the pragmatic likelihood of the different exhaustive readings. Moreover, I present a theoretical approach, which models the readings as an interaction between the semantics of the verb to know and a pragmatic principle.