[Corpora-List] ad-hoc generalization and meaning

[Rob Freeman]

And if I might try the patience of the list with just a few responses in
detail to points made by John Sowa. Where I think they might lead to broader
understanding:

On 9/14/07, John F. Sowa <sowa at bestweb.net> wrote:
>
>   ...syntax, semantics, and pragmatics address different aspects of
> language, none of them makes up for any shortcomings of the other two.


I think syntax and semantics are two sides of the same coin.

I believe we only need to start treating syntax as "incomplete", using
ad-hoc generalizations about the way words occur in context, and we will be
able to start accessing semantics in useful ways, using syntax.

> I don't think there is a lot of work which takes my position that
> > we can model semantics by making ad-hoc syntactic generalizations
> > over corpora.
>
> I am happy nobody is doing that, because it is impossible.  One might
> discover syntactic generalizations from a corpus, but it is not
> possible to learn anything about the semantics and pragmatics without
> getting further information about how people use those the sentences
> in social activities.


In contrast I believe there is a great deal of semantic/pragmatic
information in corpora, texts of all kinds. And I'm sure it would be easy to
add more. It is just a function of how much, and what kind of, context you
want to store.

If we start clustering these contexts using syntax in the way I suggest, I'm
sure we will be able to quickly start accessing all kinds of meaningful
regularities in the world.

And to consolidate with John's last post:

An important correction I would make to Rob's comments is that I
> believe semantics is the study of how language relates to the world.
> Studying a corpus is fine, but it is also necessary to relate strings
> in the corpus to the world and to life.  Without some connection to
> perception and action (even indirect), there is no semantics.


A correction to which comment, exactly? I would not disagree with any of
this.

I not only would not disagree with it, I think it can all be nicely
integrated using a corpus approach of the kind I suggest. Indeed, I think it
is implicit in the association I am making between meaning and context
(selected by syntax.) You could start by extending text corpora to speech
corpus, but really the kind of information you choose to record in a corpus
is up to you. The more context you store, the finer subtleties of meaning I
think you will be able to to model, and index using syntax.

-Rob
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.linguistlist.org/pipermail/corpora/attachments/20070915/f0ed7ad4/attachment.htm>
-------------- next part --------------
_______________________________________________
Corpora mailing list
Corpora at uib.no
http://mailman.uib.no/listinfo/corpora