In theory pure generative approaches are subject to the checks recently
outlined. However, the latest version of Chomsky's theory with which I
am familiar allows for this to be gotten around. The generative system
is allowed to overgenerate based on universal rather than specific
language-specific rules. Then, whatever does not apply to language X is
filtered out through the use of what really amounts to cannonical forms.
If something that does not fit these forms/constraints which are
language specific, it is discarded even though it has been generated.
Perhaps later versions of the theory (I think I'm going back to
Government and Binding, though there may be a bit of Logical Form here
too) deal with such theory-based problems differently.
I am reminded in this context of Bill Labov's comment on the Chomskyan
approach back in the mid 60's. He said that "they" would claim
universal applicability for their theories and then offer to prove it by
saying "Let's take any language. Hey, how about English?"
The point of all this is that when pre-conceived models take over good
theory goes out the window as I feel it has with Chomskyanism. Hope
this has been relevant.
Best,
Eileen