Re: Skinner on Language Learning

From: HARNAD Stevan (harnad@cogsci.soton.ac.uk)
Date: Tue Jun 04 1996 - 10:29:27 BST


> From: "Harte, Tom" <TH195@psy.soton.ac.uk>
> Date: Thu, 23 May 1996 20:02:20 GMT
>
> Skinnarian view - Behaviour is acquired simply through learning and
> experience.
> Like other areas of psychology Skinner explains the acquisition of
> language through reward and punishmant. The reward for saying something
> is being understood or replied to and by failing to be understood
> (punishment) the child learns through feedback of the consequences
> (trial and error). From this Skinner said that a child can acquire
> language.
> The learning theory criticises this approach using the idea of
> "negative evidence or information about which strings of words are not
> real sentances in the language to be learnt".

What is the "learning theory"? Don't you mean the evidence from actual
language learning, and the nature of our grammatical capacity, and the
best explanation (Universal Grammar, UG) of how we do it?

> This is suggesting that
> what the child hears and the feedback the child gets for what they say
> is impoverished, (the trial and error evidence the child gets is
> insufficient fot the child to derive the rules of language by trial and
> error alone). The correction of non-grammatical sentances could come
> from the parents, but it's been seen that this is not the case..

The child neither prduces nor hears enough wrong sentences. And make it
clear to kid-sib that you are talking about grammar (syntax) here, and
not about "language" as a whole.

> Children will often occupy a between stage of grammatical correctness
> and incorrectness (Using normal and past tense versions of the same
> word e.g. went and wented). If a parent isn't correcting this each time
> then how do they learn which one should become part of their language.

These mistakes are in any case not violations of UG. So they are beside
the point. The child gets and hears enough past tense errors and
corrections to learn past tens rules; even a neural net can do that.
It's the errors the child never makes, that would be needed if one were
to LEARN UG, that are a problem, and force the conclusion that the child
must already have UG in the first place.

> So if the world isn't saying to stop then something else is.
> Chomsky noticed "competance" in contradiction to behaviour. We behave
> grammatically because we have a certain capacity to do so. This
> capacity is based on a complex set of rules (Universal Grammar) which
> cannot have been learnt through trial and error. These rules are
> unconscious and inborn.

Yes, but you have to sort out the poverty of the stimulus (lack of negative
evidence) to show WHY UG must be inborn.

> Language acquisition has shown that it's virtually impossible to show
> how children could learn language unless you assume that they have a
> considerable amount of non-linguistic cognitive machinery in place
> before they start.

Repeating the statement does not support it: You have to give the
arguments and evidence; and for an A, you need to integrate it with
bigger issues, such as computation, evolution, categorisation.



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:44 GMT