Re: Fw: Intro to natural language processing
Thanks for giving me an opportunity to reply, John. It is good to see that my
crude ideas seem to have touched a nerve. I must be doing something right.
> > 2) Functionalism (focus on language as a system of contrasting
> > strings)
>
> misses the point entirely: in what sense contrasting? It
> has to be contrasting with respect to different
> functions specified in terms of the cognitive and/or
> social system: i.e., meanings construed very
> broadly. Every kind of linguistics has to do
> with contrasting strings... this is basic structure
> from the past 60 years of linguistics, nothing
> particularly generative.
I wasn't missing the point. I was making a point which you have missed.
Functionalism as a whole has missed this point: contrast in terms of meanings
is neither here nor there, it is the contrastive principle itself which
matters. Cognition is contrastive, meaning is a side effect. (So ontologies
should be regarded as a side effect of different perspectives across
structural contrasts, BTW.)
Those are pretty big claims. How do I justify them?.
Let's scan a quick history of structural contrast in linguistic theory for a
start.
You say every kind of linguistics has to do with contrasting strings. Well, in
every branch of linguistics contrasts exist, but not every branch of
linguistics promotes contrast to the status of fundamental principle. That
principle was an invention of American Structuralism as far as I know. (You
may know better.) Anyway, the point is it was specifically abandoned by
Generativism. Perhaps more than anything else that was the doctrinal shift
which defined the new discipline.
So contrast itself is a significant point of contrast between linguistic
theories, as I tried to sketch. Before Chomsky contrast as a defining
principle might have been widely accepted, but after Chomsky hardly anyone
gave it fundamental status. To its credit, Functionalism did. To that extent
Functionalism is good, and that is what I sought to emphasize.
So now we have contrast in its proper place: always there, promoted to a
fundamental principle by the American Structuralists, demoted by Chomsky in
favour of an unobservable abstraction, Universal Grammar, but retained by
Functionalists, though only as a slave to another kind of unobservable
abstraction, function.
I think that is a clearer picture.
That's what happened after Chomsky swept the contrastive principle from the
table. But how did he do it? How did Chomsky persuade a generation (except
the Functionalists) to abandon contrast as the fundamental principle of
linguistic structure? Well, he simply pointed out that the contrastive
principles of American Structuralism (the bi-uniqueness principle?) resulted
in structural representations which were "incoherent and inconsistent".
As a result one bunch of structuralists abandoned contrast and became
generativists, and another, much smaller group abandoned structure, and
became functionalists. Or perhaps the Functionalists, and Firth before them,
were always working on their own brand of contrastive significance?? The
result was the same. Structural contrast almost disappeared as a fundamental
principle of linguistics. It was an astonishing result, really. All credit to
Chomsky for the observation. His explanation was wrong, but was the path
taken by functionalism much better?
Why do I say that? Well, look at the two alternative explanations at issue:
Universal Grammar or function.
Chomsky thought (thinks) the solution to the puzzle of incoherent and
inconsistent contrastive structure was that language was really based on a
set of structural abstractions, Universal Grammar, which exist innately in
humans, and serve to select the "linguistically significant" generalizations
from all the rest. Of course this isn't reasonable. It is an absurd idea to
suggest objectivity is innate just because you don't see it!
But is function much better? Function may be an objective motivator, but what
is function? We only know it exists because it is hypothesized to motivate
contrasts in language. It is not something forced on us by observation at
all. At the end of the day thinking only about function is just another
escape into abstraction, like U.G. More a retreat from complex reality than
an explanation of it.
Let's look at what we can see. We're engineers, surely we can take it. What we
can see is that abstractions of structure, and indeed function, can be made
in terms of structural contrasts in language, but that these abstractions are
"incoherent and inconsistent".
That is the consequence of putting contrast itself back in the position of
primary motivating principle. Is it really so bad? Is a subjective model of
language (and perception) which we _can_ find worse than an objective model
of language (function or U.G.) which we _can't_ find?
Why can't we believe that this subjectivity is not a flaw in the system, that
it is fundamental to the nature of perception (and our description of
perception in language) which is based on contrast?
In that case the idea of anchoring language in functional contrast was right
(in fact, it is only to the extent that you have objective structure that you
have an objective concept of function!) But its elaboration, which abandoned
structural contrasts as a practical tool for the description of anything, was
wrong.
This is the point I was making in my comp.ai.nat-lang post, albeit lightly for
the benefit of a naive audience. It's a subtle point, hard enough to make
clear even to those well versed in the field.
What this means as a practical matter is that Functionalism has come adrift
searching for absolute abstractions of function and meaning, when in fact
function and meaning should be modelled subjectively, in terms of structural
contrasts.
Which BTW would be why you have failed, and are doomed always to fail, in your
efforts to find an absolute ontology. (Note: that is not a pessimistic
statement, what is a little depressing is that people persist in failing to
see its power.)
And yes John, these are new ideas. But that doesn't mean they should be taken
with any more salt than the standard theories. In fact it is the standard
theories which require heavy seasoning. They are, after all, the ones which
continue to fail us :-)
> Perhaps he'd also like to see some of the
> work of those who have been working in functionalist
> linguistic traditions and computation for 20 years
> or more. Then he can tell us how we are doing it wrong
> (because our systems are not getting that much better
> that fast either!
I've seen enough. Professional Functionalists spend most of their time trying
to squeeze a subjective reality into an objective description, just like the
Generativists. The only difference is they do it for categories of meaning.
So John, my ideas may be raw and unpolished, full of errors of fact even, but
is the orthodoxy really so satisfactory? 20 years with no results. Aren't you
bored? Why not look at something new? I think my arguments (and others on
similar lines) merit a more thorough exploration.
-Rob Freeman
P.S. I thought John Sowa's Oct 19th post was good:
http://grouper.ieee.org/groups/suo/email/msg12740.html
I also found it relevantly amusing that 19th century physics was held up
because Ernst Mach couldn't accept atoms were unobservable
(http://grouper.ieee.org/groups/suo/email/msg12748.html). Though when the
parallel is drawn, I'm not sure on what side of that debate I find myself :-\
Oh, and if you like Luc Steels' stuff you might like what this team is doing
up in Finland (e.g.
http://www.cis.hut.fi/~tho/publications/honkelawinter2003tr.pdf, and they're
having a conference: http://www.cis.hut.fi/AKRR05/.) I think consensus is a
good model for the emergence of lexicon. But the the consensual concept of
meaning does not explain the creation of new meaning by syntax (From memory
the best LS has is genetic algorithms selecting new syntax rules by
consensus, not quite the same thing.)