[Corpora-List] Chomsky and computational linguistics

[Rob Freeman]

On 7/31/07, maxwell at umiacs.umd.edu <maxwell at umiacs.umd.edu> wrote:
>
> > I agree Mike. I don't think Chomsky argued against observational
> adequacy.
> > Any theory must explain the data. I would say Chomsky argued against
> > observational sufficiency.
> >
> > I hope that won't now lead to an argument about the meaning of the word
> > "sufficiency".
>
> You won't get an argument from me about observational sufficiency, and in
> fact I think that's exactly right: a grammar that captures all of some set
> of observations (whether those are from a corpus, or from introspection,
> or both) is not necessarily complete nor necessarily a grammar of what a
> native speaker knows, and therefore not sufficient.


I'm glad we can agree on this Mike.

But there is a twist. You are talking about grammar_s_, plural, as an
example of observational insufficiency. I want to focus on that.

I think that was indeed the problem Chomsky saw.

Now, to Chomsky that meant something in the mind of a native speaker must
select between them.

I don't think that hypothesis panned out.

However there is another way to interpret the same data. Same data,
different conclusion.

To me the fact we get many grammars from the same set of observations
(observational insufficiency) means they are all good, and we need to keep
the observations so we can find the one we need, when we need it.

Do people see this not only keeps the corpus linguists happy, the corpus is
the only way to get all those grammars, but it also makes the machine
learning guys happy, because they can describe more structure with more
precision (more grammars, and it explains why they could never find one
complete one.)

-Rob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20070801/19ce873f/attachment.htm>
-------------- next part --------------
_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora