Diesing (1996, 1997) observes that the interpretations of object-shifted objects and non-object-shifted objects in Icelandic object shift constructions differ along lines very similar to the interpretation differences between scrambled and non-scrambled objects in e.g. German. The present paper argues that Optimality Theory has certain advantages over e.g. Minimalism in accounting for such data. This is because the interpretational differences only hold of object shift constructions: In a construction where object shift is possible, a non-object-shifted object only has one interpretation (parallel to a German non-scrambled object), but in a construction where object shift is not possible, a non-object-shifted object is ambiguous (interpretable either like a German scrambled object or like a German non-scrambled object). In other words, what matters is not just whether the object has moved, but also whether it "could have moved" (i.e. it depends on how well those competing candidates are doing which contain object-shifted objects). In Optimality Theory, such a situation can be accounted for in terms of violable constraints, and the difference between object shift and scrambling can be derived from different rankings of the same constraints.
Studies in Generative Grammar [sgg], 2001, p. 321-340