Think of it like this: Chomsky was testing the brain's language parser by handing it weird things and seeing how it reacts. Like you might do with a new programming language: what happens if I try to add strings, or divide them? Is zero true? Is the string "nil" true? Is == different from ===? Can I pass a function into a function?

The brain's language center is undocumented, so we try throwing potential sentences at it and see what works or doesn't, then try to reverse engineer what it's doing. The buffalo sentence conforms to the rules we know about word order, and can be logically explained, but somehow it fails. Finding out why is part of the reverse engineering process.

Yes... the interesting question is whether it fails because of a "rule" we're not aware of, or because it's simply too complex. The human mind is recursive, but is it simply that it only have a "stack depth" of 3 or 4 and can't parse more deeply than that?

>The human mind is recursive, but is it simply that it only have a "stack depth" of 3 or 4 and can't parse more deeply than that?

That was Chomsky and Miller's theory (although they weren't dealing with that particular example).

The interesting thing to me is that the buffalo sentence is not just grammatically correct, but semantically correct as well.

But how does that go against anything Chomsky says?

(Obviously generative grammar doesn't tell the "whole story" about language, but generally speaking, nothing tells the whole story about anything.)