[Corpora-List] Is a complete grammar possible (beyond thecorpus itself)?

[Rob Freeman]

On 9/10/07, maxwell at umiacs.umd.edu <maxwell at umiacs.umd.edu> wrote:
>
> Rob Freeman wrote:
> > Attempts to describe natural language as a formal system have been
> > uniformly unsuccessful. One reaction has been to reject all
> > formal analysis...
> >
> > My argument here has been that no-one has considered a third possibility
>
> And of course the second possibility is to improve the formal grammars.


That's actually the first possibility in my list Mike, but your point is
taken :-)

Actually, there are formal grammars which, for well-studied languages like
> English, have a _reasonably_ good coverage.  (Well, I admit that what
> counts for "reasonably good" depends on your perspective!)  At any rate,
> when the formal grammar over-generates, there are at least three potential
> explanations:
>
> (1) The grammar is wrong;
> (2) Some grammatical constraints which can plausibly be said to be
> external to the grammar itself are missing (this is the approach John Ross
> started back in the 60s, and which is still going today); or
> (3) The over-generated forms are ruled out by something extrinsic to
> grammar/ language itself (the 'colorless green ideas' explanation).
>
> If the grammar under-generates, at least (1) and (2) could be explanations
> (for (2), the explanation might be that the constraint is wrong, or just a
> weak constraint).  It may also be a lexical problem, e.g. a word missing
> from the lexicon (the "klatu verata nikto" problem).
>
> In sum, there are, it seems to me, good reasons not to give up on the
> formal grammar approach.


If you've got another 50 years to waste there are!

Symbolic grammars are just the wrong complexity.

It is like you plant the flower and try to grow the seed.

But each to his own. My argument for now is just that no-one has tried the
third approach. I'm not saying another 50 years tinkering with a-priori
feature structures doesn't sound like fun. I'm just saying we should at
least explore the possibility formal grammars are "necessarily incomplete"
descriptions of corpora, that the right way to handle language is to
generalize grammar ad-hoc from examples, as you go.

-Rob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20070910/4e23a12d/attachment.htm>
-------------- next part --------------
_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora