[Corpora-List] Moving Lexical Semantics from Alchemy to Science

[John F. Sowa]

On 1/23/2011 4:06 PM, chris brew wrote:
> ... what lexical semanticists need is to combine analyses of things like
> the Google data with careful and scholarly thinking about what ELSE one
> one can possibly know about the problems in hand.

I strongly agree.

> The contrast with alchemy is not the one I'd choose, what is
> actually needed is a combination of *real* science and *real*
> humanities-based scholarship.

My recommendation is to take seriously the following tongue-in-cheek
remark by Graeme Hirst:

    "The solution to any problem in artificial intelligence (AI) may
    be found in the writings of Wittgenstein, though the details of
    the implementation are sometimes rather sketchy."

    -- G. Hirst (2000) Context as a spurious concept, in
       Conference on Intelligent Text Processing and Computational
       Linguistics, Mexico City, pp. 273-287.

In agreeing with Graeme, I would add that all other attempts to resolve
these issues are either sketchy, alchemy, or huge volumes of statistics
that have not yet been interpreted.  (As Richard Hamming said, "The
purpose of computing is insight, not numbers.")

Before you can apply statistics to any corpus, it is necessary to
have a model that specifies what you intend to count and how to
recognize it when you see it.

If you have a POS tagger, you can find N-N compounds, but extracting
them from a corpus takes them out of the context in which they were
used.  LW emphasized meaning as use within a language game.

That raises a few more questions:  How do you recognize use?
How do you recognize a language game?  How do you recognize
which use in which game a particular N-N compound plays?

I'll admit that LW was sketchy on those issues, but at least he
was pointing in a direction that can benefit from a combination
of scieance and humanities-based schloarship.

Following is my sketchy attempt to gather a few more aspects of
science and scholarship that may be relevant, but it raises even
more questions:  http://www.jfsowa.com/pubs/rolelog.pdf

John Sowa

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