Re: cognitive != observable? (project three)

"Bill Modlin" <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...

 >
>  but we have
> > discussed this issue of reference and making-sense [to some extent]
> > in the past. Babies instinctually "babble" essentially the entire
> > set of phonemes present in language. This is built into their
> > brains at birth, and not learned after birth - as evidenced by the
> > fact that even deaf babies babble. These sounds are reinforced in
> > hearing babies by feedback from their parents/etc, and deaf babies
> > eventually stop babbling because of absence of said feedback.
> > However, their parents didn't learn these sounds from "their"
> > parents, they were initially born with this ability too - just like
> > the babies here.  There's your reference.
> 
> Hmm.  I have no quarrel with your description of what happens.  But I
> can't make sense of the last sentence... I don't understand how the
> existence of a repertoire of innate sound-making behaviors, some of
> which are reinforced, automatically is supposed to establish a reference
> for the complex mixture of sensory signals that result from those
> sounds.
> 

Well, I meant it was a possible reference w.r.t. the opposite
situation of the baby being born not having any innate knowledge of
sounds/speech/etc at all - in which case it would have nothing
whatsoever in the way of innate knowledge to start with and would have
to learn "everything" from zero.

However, in the case that there might not be anything at all on the
reception side [something I doubt anyway] other than purely naiive
learning modules/circuitry, the baby being born with an inherent
facility for babbling could possibly be seen to "train" its own
reception circuitry. If you want to take this case, then the baby
babbling sounds going out the mouth would be received by the ear, and
these sounds would then be the natural "training set" for the learning
taking place on the reception side - along with any similar
reinforcing sounds from parents/etc. So, in this possible scenario,
the baby would actually be training its own brain using its innate
babbling ability - instinctual cktry trains generic-learning cktry.

OTOH, personally I doubt the foregoing is the entire case, but rather
there is also some circuitry on the receptor side which is in some way
tuned to the babbling out the mouth side. Somehow it seems unlikely
that evolution would just build one side [babbling output] and not
build something complemetary on the other side ["matching" receptor
cktry of some sort], but you never know. I'm simply making up possible
and testable hypotheses. Neuroscience will figure it out one of these
days.

Regards "... supposed to establish a reference for the complex mixture
of sensory signals that result from those sounds ..." - the entire
first two years of baby life are building up one cognitive structure
upon another. Of course, Piaget wrote 50 or so books about the
progressive development of cognitive structures in the first few
years. You have to start somewhere - and that is with the born-with
instincts. Everything else is built on that.
==================


> I suppose it is possible that hearing babies could be born with an
> innate ability to distinguish and discriminate the sounds that they make
> while babbling, and don't therefore have to learn to hear them.  I tend
> to doubt it, but even if in that special case there was some innate
> special processing, there are still a lot more perceptual
> discriminations to be learned...


That's what years 2 thru 5 of baby life are for. Year 1 is just the
start.
=================

> >
> > However, in line with what you say below .... what I've said above
> > is just the "starting" point towards facility in language in
> > humans, not the total solution. If your AI starts totally naiively
> > without the reference given, then it will somehow have to develop
> > it on its own [in total absense of anything built-in to compare
> > against] - and out of the confusing buzz of randomly-appearing
> > external signals.
> 
> If I want my AI to learn to speak, it would probably be a good idea to
> build in babbling (some automatic tendency to produce behaviors which
> could then be reinforced) to help it get started.  I still can't see how
> that helps it learn to hear.  I can't imagine how you could find
> anything to "compare" in the internal structure of an innate babbling
> mechanism and any of the raw auditory signals that are available.  Where
> are you seeing a reference?
> 

As above, the phonemes present in the babbling are repeated again and
again and again. The baby is essentially "selecting" them for
presentation to its own hearing apparatus.