The problem with the Chomsky nonsense about how the syntax of a language can't be learned from input (so much, so fast, with such bad data!), so it must be innate, is that they are not proving that the input is insufficient. They're just claiming it. They just say that they don't see how we could learn so much from so little, so it can't be done, so it isn't learned, it's inborn.

That's just the creationist argument: I can't see any way that something as complex as a human could emerge randomly, so it must have been assembled deliberately by a super-human creator. There's no other way.

Well, no. The thing you're missing here is that there might be algorithms that can do more than you realize.

If the Chomskians proved that the information to be learned simply didn't exist in the input--not that they didn't see it, but using information theoretic methods proved that it could not be there, then the learning algorithm wouldn't matter. It would not be possible for any algorithm.

But unless linguists prove the needed info isn't there (which is probably not possible using our limited knowledge of information theory), the ockham's razor hypothesis for how kids end up speaking the exact dialect they're surrounded with is that they possess algorithms capable of extracting the information from their surroundings.

Declaring such learning impossible because it doesn't look possible is like declaring human evolution impossible for the same reason. We're just beginning to discover that there are spectacularly powerful algorithms capable of doing things we never imagined.

Look, what you say is roughly my point of view, and I'm playing the devil's advocate here, but you didn't actually respond to the article, but rather to some straw-man representation of "the Chomsky nonsense", to use your words. I'll try to elaborate the claims in the article a bit (and I'm probably overly simplifying it):

when you have a sentence like "I’ll play with this red ball and you can play with that one", one can ask whether "one" refers to "ball" or "red ball". The sentence doesn't give you information as to which option is true. But, say there are two kids, Chris, playing with a red ball and Max, playing with a blue ball, and someone says "Chris has a red ball but Max doesn’t have one." In that case, they can learn that 'one' refers to 'red ball' and not 'ball'. Their argument is that for a language learner (read: baby) to learn correct grammar, he has to encounter such situations, and yet, such situations are rarer than the rate of grammatical errors in similar uses of the word 'one' is higher than the rate of occurrence of scenarios where you could learn the distinction (this is not an assertion, it's backed by analysis of corpora of parents talking to their babies). And yet, their experiments show in a somewhat roundabout manner that the babies treat 'one' as referring the 'red ball' and not 'ball'. Their conclusion is that this means that out of the space of possible grammars, babies didn't consider the ones where 'one' refers to 'ball' and not 'red ball' - there's an innate preference towards those.

While I can poke some holes in the argument (mostly because of the implied assumption that the mental representation of grammar is in a generative grammar), your response read too much like http://xkcd.com/793/

Well, that was an hour spent. Thanks!

Or it could be that babies learn to chunk noun phrases as units from hearing other types of speech. The corpora you mention only disprove one mechanism of acquiring antecedent skills, not prove that it is innate. Conflating the two is, in fact, typical Chomsky nonsense.