Re: Chomsky's theories

From: Ralph Hartley (hartley_at_aic.nrl.navy.mil)
Date: 11/10/04 

Date: Wed, 10 Nov 2004 16:59:50 -0500

Eray Ozkural exa wrote:
> Ralph Hartley <hartley@aic.nrl.navy.mil> wrote:

>>Chomsky's linguistic theories have
>>turned out to be, in many fundamental respects, correct.
>>
>>His most important hypothesis was that humans have an innate capacity for
>>grammar, that controls (but is modulated by) the range of structures used
>>in natural languages.
>
> It is debatable that the range of structures used in natural languages
> are trivially learnt.

What do you mean by "trivially learnt"? Do you think they are, or not? What
difference would it make? Anyway, almost anything is debatable, but that
proves nothing.

The songs of some birds are genetically determined, but don't develop in a
bird that does not hear its own species' song in the nest. You might think
that they are learning it, but they can't learn any *other* song. The
theory is that language is like that.

> It is obvious that

Far *more* obvious things have turned out to be wrong before.

> a communication language, which encodes the same
> kinds of things (e.g. commonsense meaning), using identical or similar
> I/O constraints, and identical computational capacity (for learning),
> will result in very similar grammatical forms.

Signing has very *different* I/O constraints from speech. You would expect
it to have very different grammatical forms, but it doesn't.

> The above experiments only mean that humans are intelligent enough to
> develop languages, using certain criteria of optimality

Apparently, only children under six are smart enough, since adults or older
children *never* develop languages, no matter how much they need one. Most
ways of measuring intelligence (learning, problem solving, etc.) seem to
show that adults and older children are *smarter* than younger children.

The claim that human language is optimal in any sense is pretty strong as
well. To me it seems to be just *barely* good enough. The common structures
of human language don't appear to be particularly *unique* among all
possible such structures either.

> I am personally looking for a broad computationalist answer instead of
> taking the Chomskyan school seriously. The answer is in machine
> learning, in my opinion.

I know enough about machine learning to know that learning from scratch is
*much* harder than learning a variation of a predetermined pattern.

> If you believe that Chomsky's theory has any merit, then it is likely
> that you will think that the experiments mentioned show that he was
> right.

Actually, I think they show that even though I did *not* initially believe
his theory had much merit. If you mean the theory should be dismissed out
of hand, then you aren't talking science.

I don't claim that the evidence is water tight, only that it is strong.
That's the best you can expect at this stage.

> The challenge I will put up for the syntax-centric theorist could be
> quite simple. I wonder many bits this hypothesized universal grammar,
> but never really shown directly, takes.

How many bits does the human hand take? I hope you agree that is
genetically determined, but I'm sure you don't know how many bits it takes.

If I had to *guess*, I would say low thousands, but so mixed in with the
rest of brain function that no meaningful answer is possible. Depending on
  how you count, the answer may differ by an order of magnitude. Most of
biology is like that, some bits contribute to many functions. Nothing is
broken into neat subroutines.

The human genome is about 3Gbits.

> Chomsky's complete neglection of pragmatics, which is the real issue at stake here

You have to start somewhere. Limiting consideration to syntax, which is
easier to quantify, was more productive than I would have predicted.

> Neither can these theories explain what a programming language,
> mathematical language, musical language or a graphical language is.

So? They don't explain what gravity is either. Why would you expect them to?

If you know of any experimental evidence *against* the theory, let me know.
"It obviously has no merit" is an argument that is itself without merit.

Ralph Hartley

Relevant Pages

  • language trainer - kind of thing
    ... recently I've started learning yet another foreign language ... I'm learning Spanish now. ... with comprehensive examples (I managed to find a great Spanish grammar ... repeat them often with various modifications like using different verbs, ...
    (sci.lang)
  • Re: Practicing by playing songs
    ... "I do notice that my playing is improving". ... It's like learning a language. ... The "grammar first" approach usually ends up with the student knowing "about" the language, but not able to speak or understand it very well. ... A better approach is to learn how to say some things by ear, and after that learn the grammar that explains how the sentences are put together. ...
    (alt.guitar.beginner)
  • Re: Strong AI Thesis (No Chinese room, I promise)
    ... with the idea of a generic learning system at work. ... make vision work better, and language work better, etc. ... Logic, mathematics, and scientific method ology have ... recognized the limitations which linguistic practices im pose on ...
    (comp.ai.philosophy)
  • Learning a language
    ... I am having the pleasure of learning a third language, ... English, Pidgin, Latin, and who knows what else. ... noun, verb, adjective, etc. so the complex grammar of English is ...
    (sci.lang)
  • Re: Universal grammar
    ... When I learned language, one focused much on grammar and spelling, but ... It is easy to slip into thinking this is a learning issue. ... with is not so much that we learn language from examples. ... analogize in bulk from your own language) all the forms in that set. ...
    (sci.lang)