Children don't know how to repeat.

I've written about it before - it's my firm and considered opinion that the idea of trying to learn "like a child" is contrary to all the evidence and completely counterscientific.

It's something that I used to believe in, although I thought it would only work in an intensive situation, and that trying to force it into a short format for night classes or as part of a high school curriculum was doomed to failure.  But that was before I studied language at university.  The magic of immersion is pretty alluring, right up until you see the cold hard facts behind it.
Now don't get me wrong, it's impossible to set up a large scale double-blind study on child-rearing, so most of the facts are more tepid and slightly crumbly than cold and hard, but they're maybe the best we're going to get.

What research there has been is pretty much ad hoc - all we really have is a series of case studies performed by (mostly female) academic linguists as they bring their own children up.  The main finding, as my university teachers told me, was that children cannot be corrected.  They showed us several transcriptions of attempted corrections by the mums, but the most common occurrance was for the children to repeat the same thing they'd just said, "error" and all.  Don't believe me?  Have a look at this video of a parent from YouTube...

Why can't this kid say banana?  Because he hasn't learnt the word yet.  Simple.  But surely he should be able to repeat it when he hears it...?

Well no, because he hasn't learnt the fundamental components of the word banana.  A word is not a fundamental unit -- it is composed of syllables built and combined within certain constraints that vary from language to language.  "Banana" happens to be a very unusual word in English, so the rules that govern its construction won't usually be learned until fairly late on.  The traditional conclusion is that a child cannot repeat something that they wouldn't produce spontaneously; that the child develops a "theory" of language which is constantly refined by input, and that every utterence is realised in accordance with this theoretical grammar.

But people kept telling me that I didn't know what I was talking about, because I don't have kids myself.  They have kids, and they corrected them.  Why does this sensation persist?

Well have a look at this talk by MIT researcher Deb Roy.  He's a machine intelligence researcher rather than a linguist, but trying to mimic natural language is a major theme in machine intelligence.


At 4:20 onwards we get to hear Deb's son slowly progress from "gaga" to "water".  Interestingly, you can see a period of instability - he doesn't switch immediately to "water" and seems to revert to "gaga" for a while.  If there is a perception among parents that children can be corrected, it may well be because there is a zone where the child's grammar accepts both possibilities, and at this point the child presumably can be corrected, because he can spontaneously use both the incorrect and correct forms.

He goes on to point out the patterns of parent-child interaction:

And what we found was this curious phenomena, that caregiver speech would systematically dip to a minimum, making language as simple as possible, and then slowly ascend back up in complexity. And the amazing thing was that bounce, that dip, lined up almost preciselywith when each word was born -- word after word, systematically. So it appears that all three primary caregivers -- myself, my wife and our nanny -- were systematically and, I would think, subconsciously restructuring our language to meet him at the birth of a word and bring him gently into more complex language. 

So the natural teaching process appears to occur as and when the child is ready for the language feature in question, without the caregiver ever knowing they're doing it.

The parent's perception of correction can most likely be explained thus:
The parent attempts correction frequently.
The child generally rejects the correction.
BUT
The child accepts the correction a rare few times, when they're in the unstable zone between the incorrect and the correct zone.
These few successful instances are more significant to the parent than the many unsuccessful instances, biasing the parent's memories.

Whatever, children learn whether we consciously try to teach them or not.
But children will not be able to repeat "what is your name" until they have learnt question word order, possession, "what" and "name".  An adult, however, will be able to parrot them without any access to the underlying concepts whatsoever.  A child can only learn these concepts and structures through exposure as this is our only "interface" with the infant brain, but we can get direct access to an adult brain through the language the adult already has.  It is pretty trivial to demonstrate that any successful adult learner does indeed think about a new language in the abstract, regardless of the medium of instruction: just ask them a question.

The adult learner attempting to "learn like a child" will be relying on higher-order reasoning, but the immersive environment does little to prepare the material for a higher-order approach. Conscious instruction, with native language input, gives the opportunity to do it right.

(Which is not to say that it guarantees to do it right, but that's a matter of methodology....)