[Corpora-List] Chomsky and computational linguistics

Sun Aug 5 07:56:06 UTC 2007

Isn't it about time this conversation went off the list,

Adam Kilgarriff

-----Original Message-----
From: corpora-bounces at uib.no [mailto:corpora-bounces at uib.no] On Behalf Of
Rob Freeman
Sent: 05 August 2007 01:33
To: Anil Kumar Singh; CORPORA at UIB.NO
Subject: Re: [Corpora-List] Chomsky and computational linguistics

Anil,

You seem to have missed my point in this discussion. Which was...:

Chomsky failed at "the task of generating all and only the grammatical
sentences of a language."

But he was tremendously successful in pointing out that "observable language
can't be abstracted satisfactorily." 

Unfortunately the net result of his success has been to polarize theoretical
linguistics into antagonistic factions which either reject structural
descriptions entirely, or believe such descriptions must be innate. 

That both were originally just two possible reactions to the idea that
"observable language can't be abstracted satisfactorily" has been forgotten
(Syd Lamb says his solution that language elements be combined
"non-linearly" was ignored.) 

Others have simply forgotten that "observable language cannot be abstracted
satisfactorily." In many ways that is the most common result, especially
among engineers. (This may indeed be an area where theory can inform
engineering.) 

In particular no one has considered the equally plausible conclusion that
structural descriptions of language are necessarily multiple, and selected
by context.

-Rob

On 8/4/07, Anil Kumar Singh <anil at research.iiit.ac.in> wrote:

There seems to be a widespread explicit or implicit assumption among many
linguists as well as among people from fields which have some overlap with
linguistics that Chomsky has not only failed but has refused to accept his
failure. Well, I am not really a *linguist* and I am interested in cognitive
and statistical approaches. Also, as a computational linguist, I am using
corpus all the time. But I just can't see why one should say that Chomsky
has failed. 

- During the last fifty years or so, he has done so many things that it's
impossible to say that he has failed in all that he did

- Even in the narrow sense, I don't think he has failed, because as Mike has
pointed out, the central idea was innateness and Universal Grammar, which
has been quite a success 

- As another example, I personally think the idea of autonomy of syntax and
semantics is correct and it will be proved so in the future. I can say more
on this, but may be later...

- All kinds of people have taken something from the Chomsky branch of
linguistics, e.g. cognitivists. Even computer scientists.

- Just try to imagine what linguistics would have been had behaviorism
dominated the field

- It's really not correct to say that Chomsky has refused to 'accept that he
has failed'. I don't remember the source or the exact words now (someone on
this list surely would) but he had explicitly said that he doesn't claim to
know what exactly is the correct solution. He had written that if at all we
some day find the correct solution*, most probably the (specific) solution
he is suggesting will turn out to be wrong. 

  * Which we might not: his famous spider and the web example. Like the
spider, we may have this great skill of language but we may never get to
know how exactly we use it.

- His churning out new theories every decade shows that he never claimed to
have found the correct solution. He just claimed to be trying to get nearer
to the solution.

- I think it's unfair to just look at his specific theories and based on
their (partial) failure claim that he has failed. What he has been doing is
much more than just proposing some new grammatical theories. 

- I think, on the whole, he has succeeded more than he has failed. Even his
failures (if they are that) have added to our understanding of how language
works.

- A lot of his presumed failure has to do with the kind of goals he had set
for himself and for linguistics. He wanted to do linguistics the way
physicists do physics. No wonder he considered semantics to be out of the
scope. Can anyone really claim that we can (even after his 'failure' and
some others' 'non-failure') talk about semantics in the way physicists talk
about physics? I don't think we should restrict ourselves to physics-like
study and so I am not averse to speculating about semantics. I think the
best work on semantics (including computational) is at the same level (on
the scale of being scientific) as the political work of Chomsky. And that is
quite alright because it: 

    - is a sincere attempt to find the truth
    - is rigorous
    - tries to stick to really scientific methods as far as possible (not
always possible)
    - may be practically useful

- That some ways of inquiry were 'blocked' is as much as a fault of others
as his. Others could have tried new ways irrespective of what he said. That
a lot (or all) of them did not is something to do with the way society
works, not just about his views. 

- Language is so complex and so important a part of our psychology (and
philosophy and social behavior and politics and ...) that, as they say in
computational theory, if this problem is solved, all the problems will be
solved. Why should there be any surprise that Chomsky, or anyone else for
that matter, has failed to come up with a complete and correct solution. I,
for one, am extremely thankful that the mysteries of language haven't been
all solved and am hopeful that they won't be: at least in the near future. 

- As some others have pointed out, what is the right way and what is not may
depend on your purpose. If I just want to automatically identify the
language of a document and a purely statistical method (learning from a
small corpus) gives me the right answer almost always, statistics is the
right way for me for this purpose. But that doesn't necessarily mean that
Chomsky has been proved all wrong. 

- Finally, my favorite example (from Chemistry): Dalton, in his formulation
of the atomic theory, confused atoms and molecules (which Avogadro later
pointed out). Did Dalton fail and Avogadro succeed? In an extremely narrow
sense, yes. Otherwise, not really. 

Anil Kumar Singh 

On 7/28/07, Terry < tmorpheme at hotmail.com <mailto:tmorpheme at hotmail.com> >
wrote:

You seem to be saying that because one physicist has failed, all physicists
should give up. I don't think that follows. But it is perfectly possible for
a physicist to concede defeat without shame. Stephen Hawking provides an 
example. And if you read a physicist like Brian Greene he does say
explicitly, for example, that Einstein failed in coming up with a unified
field theory. Must we wait until Chomsky dies before declaring his project a

failure?

At any rate, I think it is time that people started saying this kind of
thing about Chomsky. He failed, but for whatever reasons of ego
preservation, he has never admitted it.

Terry

-----Original Message-----
From: Mike Maxwell [mailto:maxwell at umiacs.umd.edu]
Sent: Saturday, July 28, 2007 1:06 AM 
To: Terry
Subject: Re: [Corpora-List] Chomsky and computational linguistics 

Terry wrote:
> I wonder if someone could clear up a small point. I understand that in
> Syntactic Structures, Chomsky set himself the task of generating all and
> only the grammatical sentences of a language. 
>
> Presumably, in the intervening 50 years or so since that work, he has
never
> succeeded. But has he ever said: "I have failed in this task"? Surely, at
a
> certain point, this is what a scientist would do. 

You could argue on those grounds that physicists should give up because
in a century of trying, they haven't succeeded in finding the ultimate
theory of the constituents of matter.

More to the point, one of the issues is that linguists are trying to 
reverse engineer nature.  When all you have is a black box, and you're
trying to figure out what's inside it, it's not surprising if it turns
out to be difficult.

This is going to date me :-(.  When I was about ten, I took apart a 
wristwatch, and basically figured out how it worked.  For my next
project, I decided to take apart a TV to see how it worked.  Inside the
TV were hundreds of pieces--things which I now know to have been
resistors, capacitors, tubes, etc.  Have you ever taken a resistor or 
capacitor apart to see how it works?  Unlike a watch, there are no
moving parts.  I was defeated.  Now at that point, I could have
re-invented the theory of electricity, or I could have read a book.  I
did the latter.  Unfortunately, God hasn't published His book on how the 
brain works.  So linguists are left trying to re-invent the theory of
electricity, and it has not turned out to be simple.

On another level, "all and only" is the level of observational adequacy.
  On the way to that, a number of puzzling properties have turned up
(island phenomena, an apparent distinction between long-distance vs.
local movement, preposition stranding vs. pied piping of prepositions,
etc. etc.).  While it would be possible to handle some of these on an ad 
hoc basis (some might claim that the principles-and-parameters approach
was just that), it would be interesting to know whether these facts
might follow from some deeper theory.  This is the search for
explanatory adequacy, and it is that search that Chomsky and others like 
him have chosen, leaving to others the task of observational adequacy
(and maybe descriptive adequacy).

You can say that this search for deeper theories before we have an
observationally adequate treatment is the wrong way to go, but in the 
end I think there's room for both.  And (possibly contrary to some other
opinions expressed in this thread, if I understand them), I don't think
Chomsky has argued that no one should work on observational adequacy. 
On the contrary, data collection and observationally adequate accounts
of that data have served on several occasions as turning points in
Chomsky-style linguistics.  One example of that is the discovery of
parasitic gaps.  If I'm recalling correctly (and I'm probably not :-(, 
but I'm sure someone will correct me), Elisabeth Engdahl described them
from a construction grammar perspective (which was more or less
observationally adequate).  Chomsky's reaction was that these were 
fascinating, but rather than being a special construction that people
learn (on the basis of extremely limited data), they ought to fall out
from other principles (explanatory adequacy).

    Mike Maxwell
    CASL/ U MD 

_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora 

_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20070805/6e673f20/attachment.htm>
-------------- next part --------------
_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora