In this talk, I present new empirical data showing that the expression of the sameness relation (with the use of, for example, symmetric be (see also Cumming 2008, Percus & Sharvit 2014), internal same, etc) in de dicto attitude reports correlates with the absence of corresponding de re attitude reports.
Based on this, I propose a new characterization for the semantics of sameness expressions in natural languages: essentially, such sentences as 'A is B (and B is A)' or 'A and B are the same' mean that the contextually salient properties of the res named A in a certain world hold for the res named B and vice versa.
The current proposal can very naturally account for (i) the behavior of sameness expressions in attitude reports, (ii) the scope taking behavior of internal same (see also Barker 2007), and (iii) the island effects of internal same (see also Hardt and Mikkelsen 2014). Moreover, the current analysis makes an interesting new prediction: the sameness relation expressed in natural languages is not transitive.