Re: [Corpora-List] QM analogy and grammatical incompleteness

From: Rob Freeman (lists@chaoticlanguage.com)
Date: Fri Dec 23 2005 - 01:02:09 MET

  • Next message: Liliana Mamani Sanchez: "Re: [Corpora-List] Unsupervised machine learning"

    On Friday 23 December 2005 06:07, Ken Litkowski wrote:
    > As long as we're on a QM trip, how about an excursion into QCD (quantum
    > chromodynamics), where an important constituent is the particles that
    > mediate various forces. While we may think of semantic primitives
    > (e.g., a la Wierzbicka) as analogous to protons and neutrons, etc., what
    > about primitives of semantic relations, as reified substantially via
    > prepositions? Or, is this a stretch?

    I think we'd better protect the list from QCD for now Ken, but sticking
    narrowly with grammar for the minute, I do think it is very useful to imagine
    these indeterminate, distributional classes combining to form new classes.

    The combination fixes the indeterminacy in a way. Prepositions have
    distributions in terms of word associations too. Put in adposition with the
    distribution of some content word you can think of the distribution of the
    preposition "filtering" the distribution of the content word to resolve some
    of the ambiguity/indeterminacy.

    It is better to do it this way. If you try to resolve the content word down to
    some grammatical label before it is combined with the preposition you are
    bound to lose information (all other classes become fuzzy.) It doesn't make
    any sense to label the content word until you know what it will be combined
    with.

    You could of course try to project out all possible classes from the
    distribution before you make any combinations. But just how many ways could
    the distribution of that content word be resolved into classes?

    In a sense the very indeterminacy of the distribution gives it a new power.
    That one distribution can represent all kinds of classes, to be projected out
    using the right combination of words.

    For the sake of leaving the general grammatical representation of the language
    in its raw form as a corpus (perhaps indexed as a distribution), you can have
    an almost infinite number of resolutions in terms of grammar, and meaning,
    projected out by appropriate combinations of words when you want them.

    -Rob



    This archive was generated by hypermail 2b29 : Fri Dec 23 2005 - 03:14:07 MET