[Corpora-List] ad-hoc generalization and meaning

[John F. Sowa]

Rob,

 > What problems caused Montague grammar to be rejected?

It is just too rigid in both syntax and semantics.

As I said in my previous notes, even the best parsers
in existence today can only generate completely correct
parses for about half the sentences of a well edited
document in a genre for which the parser has been well
tested and trained.

Of those 50% of sentences that can be parsed correctly,
Montague semantics is lucky to get a very tiny percentage
of them correctly mapped to a logical formula.

That does not make it useful for practical applications.

 > Would they not be solved if grammatical patterns were seen
 > as the result of ad-hoc generalization over a corpus, and
 > not complete in themselves?

The primary goal of Montague grammar is to support the semantics
-- i.e., translating a sentence to a formula in logic.  The
grammar is already more robust than the semantics.  If you
improved the number of successful parses for the grammar, that
would do little or nothing to improve its percentage of correct
logical formulas generated as the final result.

John


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