The availability of the factive inference in Turkish depends on a) the syntax of the embedded clause, and b) the position of new information focus. Given a predicate like "bil-" (which, in some contexts translates the English "know"), tensed embedded clauses are never factive (1), nominalizations with embedded focus are non-factive (2), and nominalizations with matrix focus are factive (3).
1) Tunc Ankara'dayim diye biliyor
Tunc I'm~in~Ankara PRT knows
Tunc has the justified belief that I'm in Ankara.
2) Tunc [Ankara'da]_foc oldugumu biliyor.
Tunc in~Ankara I~am.NMZ knows
Tunc has the justified belief that I'm in Ankara.
3) Tunc Ankara'da oldugumu [biliyor]_foc.
Tunc in~Ankara I~am.NMZ knows
Tunc knows that I'm in Ankara.
My first goal in this talk is to describe this complex empirical landscape. What are the factors that determine the availability of the factive inference? What remains when a predicate like "bil-" is defactivized? I argue that justified belief remains---which is strictly stronger than plain belief. My second goal is to sketch out an account of the factivity alternation. Predicates that participate in the factivity alternation denote non-factive attitude relations that are de re about particular situations. The factive inference is then generated in the semantic composition.