Séminaire de Recherche en Linguistique

Factive verbs (regret, remember) select CPs that are presupposed to be true; non-factive verbs (believe, want) don’t (Kiparsky & Kiparsky (K&K) 1971). Long-distance extraction out of factive CPs create Weak Islands (WI), with only argument extraction being possible (Rizzi 1990, a.o). In addition to WIs, Romance and Balkan factive constructions may also involve Strong Islands (SI, when both arguments and adjuncts are banned for extraction). The traditional approach in the literature is that the embedded CP is actually different in the case of factive constructions (K&K 1971, Rizzi 1990, de Cuba & Ürögdi 2009, Haegeman & Ürögdi 2010, a.o). On the basis of Romance (French (Fr.), Italian (It.)) and Balkan (Torlakian Serbian (TS), Croatian (Cr.), Bulgarian (Bg.), Modern Greek (MG)), we claim that this approach is basically correct, but with the added detail that the matrix predicate’s structure is crucial too. More precisely, we claim that three features are responsible for the island effects observed; normally these features are spelled out as a complementizer, but sometimes they can be split between the matrix verb and the complementizer, with consequences for lexicalization.

In other words our analysis accounts for factive islands in Romance and Balkan. In both cases, the complementizer is responsible for the blocking effect, but features of the complementizer-fseq can also be absorbed on the matrix verb, with lexicalization effects on the complementizer.