One of the maxims of universal grammar is that children's language acquisition indicates the existence of a genetically preprogrammed language faculty. Because a child cannot master certain complex tasks, yet manages language - so the argument seems to go - there must be something else, something innate that enables said child to speak and understand sentences from an early age.

This seems a fairly convoluted conclusion. Considering that millions of children acquire languages every year, and that a sizeable subset of these children learns multiple languages with great ease, the more fitting conclusion seems to be: "language is not too complex for children to acquire". (For, at the risk of sounding painfully obvious, they are children and they do indeed acquire it.)

The nativist deduction seems a non sequitur.

What exactly is it about language that should be too complex for children? How would one define this complexity? Are there any neurological reasons why this complexity should not be grasped by a child's growing brain?

  • 1
    It is a non sequitur. UG is a hypothesis that is striking because it's so counter-intuitive. It certainly would be nice if it were true. But there is no evidence except negative evidence (i.e, "Very very smart people, like me for instance, haven't been able to figure all this syntax out, so how could a mere child? It must be already there in the brain.")
    – jlawler
    Mar 29, 2014 at 16:35
  • 4
    The usual argument that's given is the 'Poverty of the Stimulus' - the idea that the evidence a child is exposed to isn't sufficient for them to draw the conclusions they do about the grammar, hence there must be something else, some additional body of knowledge constraining the child's hypothesising. Before somebody pounces, i'm not endorsing the position, just clarifying the argument. You can find a recent overview of the state of the PoS argument here: isites.harvard.edu/fs/docs/icb.topic1327223.files/…
    – P Elliott
    Mar 29, 2014 at 18:00
  • 2
    Construction Grammar is a good alternative, and so are the many other flavors of cognitive grammars. They're more interested in data than speculation, or at least more circumspect in keeping them separate.
    – jlawler
    Mar 29, 2014 at 20:37
  • 1
    And to keep things balanced here's a paper by David Adger laying out some weaknesses and drawbacks of constructionist approaches: ling.auf.net/lingbuzz/001675/current.pdf (the host is down right now incidentally, but should be back up soon).
    – P Elliott
    Mar 29, 2014 at 23:11
  • 1
    I don't really feel that this question is easily answerable - it's also an extremely loaded question, in that it's clear where the questioner's sympathies lie. We can only really know the answer if we have a concrete acquisition algorithm to test against the linguistic data. The question is how many priors need to be built into the algorithm such that the grammar can be acquired on the basis of the evidence. In the absence of that, this is all empty speculation.
    – P Elliott
    Mar 29, 2014 at 23:15

1 Answer 1


You misstate the crux of the so called "poverty of the stimulus" argument. It does not say that language is too difficult for children because it obviously is not. The point is that even with remarkably impoverished stimulus (with some very few extremes like feral children being the exception), children seem to acquire syntactic structures for which they simply did not receive enough input to form meaningful generalizations.

This argument has been attacked from many angles, some from misunderstanding but others, in my view completely convincing.

  1. A lot of argumentation is about whether children don't really receive a lot more implicit instruction in grammaticality. And that's certainly true. But the POtS argument does not dispute that. It merely says that even with almost no such stimulus, children still acquire those structures. It also says that the sort of linguistic nurturing we take for granted is not at all common across cultures which still does not inhibit acquisition. So this is the one part of the argument that stands.

  2. A more successful attack is that on 'learnability'. This argument says that for POtS to stand, we would need some objective concept of learnability for each structure. Something that would quantify how much exposure to natural language a person needs to learn something. And not even a hint of that is offered by POtS proponents. Moreover, many of the claims on exposure have been challenged by corpus analysis. Of course, just because we can't specify learnability in a falsifiable way, it does not mean it does not exist. Samson wrote a whole book punching holes in this concept.

  3. Another challenge (my favourite) is that universal grammar does not really represent much of what language is. It's a handful of rules (more or less disputed by linguists) of limited relevance to communication. Let's just look at the awesome amounts of language a child has to learn that UG has nothing to say about. Thousands of words, each with incredibly complex usage rules. A child has to develop judgements about register, social appropriatness, dialect, etc. So there's a lot to learn and it's obviously not too much. But if you were to write it all down, it would fill volumes. So why are a few relatively simple structures such a standout? Isn't more likely that we're missing something about how they're acquired than needing to posit a whole new universe of hard-to-prove innate structures?

  4. But the killing blow was dealt by Thomasello and researchers in his tradition. Most convincingly, research by Ewa Dabrowska showed that children really don't acquire the troublesome stuctures in the same way that Universal Grammarians imagine. They acquire things like recursion only partially and are not at all able to generalise to more complex instances. In fact, this variability persists into adulthood. Ability to process complex recursions varies by how much exposure one has had to language in which they often appear.

So the answer to the question "What exactly is it about language that should be too complex for children?" is that on the surface almost everything should be too complex but in reality almost nothing is. The only things that are too complex, are those that not everyone can learn or not learn equally well. But those depend on a combination of availability of input and generalized (rather than very specific) ability.

  • (For what it is worth, I still don't even remotely understand the poverty of stimulus argument. People literally use thousands of words per day. Assuming an infant has a parent, said infant would have heard hundreds of thousands of sentences before he would start speaking. Whence the poverty?)
    – player.mdl
    Mar 30, 2014 at 16:12
  • @player.mdl Knowledge of language doesn't merely consist of lexical knowledge. We've managed to train a dog to understand over 2000 words after all (newscientist.com/article/…). The question is really how an infant can hypothesise a grammar such that it can produce complex sentences that it has never heard before. Language learning doesn't consist of learning sentences by rote, after all.
    – P Elliott
    Mar 30, 2014 at 16:26
  • @player.mdl If you don't understand the basic argument, i'd suggest reading the article i linked to in my comment under your question!
    – P Elliott
    Mar 30, 2014 at 16:27
  • I could have phrased the previous comment better. I understand it. I don't at all see any sense in it ;)
    – player.mdl
    Mar 30, 2014 at 16:39
  • 1
    The thing is that it's pretty difficult to actually give a concrete account of how children acquire language so rapidly using only domain general mechanisms. Until someone comes up with a concrete, formal learning algorithm with no built-in linguistic-specific priors the acquisition process will still be a puzzle. To my knowledge, such an algorithm doesn't exist.
    – P Elliott
    Mar 30, 2014 at 16:52

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.