The author seems to be hung up on a distinction between two representations and trying to argue that they cannot be the same thing, despite the fact that we hav

The author seems to be hung up on a distinction between two representations and trying to argue that they cannot be the same thing, despite the fact that we have abundant evidence that both are readily interconvertible. Now, I would agree that a neural net that converges to a grammar might not be a grammar, but at that point we would seem to be missing the forest for the trees.