[Fwd: Re: [developers] processing of lexical rules]
Emily M. Bender
ebender at u.washington.edu
Fri Feb 11 19:06:40 CET 2005
Hi Berthold,
It's true that the cases I am most concerned with involve a single
linguist or team of linguists doing both morphological and syntactic
work. (I'm working towards software tools to aid field linguists
in language documentation.)
If you're using someone else's morphological analyzer, I guess
it would be relatively common to find that there are things in the
output which are irrelevant from the grammar's point of view.
Adding a mapping from the output you get to the input you want
seems like a sensible approach in that case. Did it in fact turn
out to be cumbersome with Morphix?
Emily
On Thu, Feb 10, 2005 at 02:16:41PM +0100, Berthold Crysmann wrote:
> I am not sure that will necessarily give you efficient processing,
> unless you can control the morphological component:
> e.g. for an adjective like anderen, I get 7 distinct analyses from the
> Xerox German demo, owing to syncretism in German:
>
> anderen ander+PAdj+Indef+Fem+Sg+DatGen+Wk
> anderen ander+PAdj+Indef+Neut+Sg+Gen
> anderen ander+PAdj+Indef+Neut+Sg+Dat+Wk
> anderen ander+PAdj+Indef+Masc+Sg+AccGen
> anderen ander+PAdj+Indef+Masc+Sg+Dat+Wk
> anderen ander+PAdj+Indef+MFN+Pl+Dat
> anderen ander+PAdj+Indef+MFN+Pl+NomAccGen+Wk
>
> In the LKB, I can compactly represent all these readings as one type.
> Either it will be necessary to map all these readings to an
> underspecified representation, as we used to do with Morphix output, or
> one would have to modify the implementation of the finite-state
> grammars to better reflect the possibilities offered by type
> abstraction. The second possibility may also depend on licensing issues,
> so it might not be possible at all.
>
> Berthold
More information about the developers
mailing list