[Fwd: Re: [developers] processing of lexical rules]
Jeff Good
good at eva.mpg.de
Fri Feb 18 15:11:57 CET 2005
Hello everyone,
I've been following the discussion of morphology in the LKB over the
last week or so and now that Emily has brought up the Montage project,
I thought I would share some of my current thinking on how Montage's
"morphophonology" component would work (much of the stuff that's been
discussed Emily and I are calling "morphophonology" to emphasize an
idea that we will use one system to "unpack" complex "squished"
phonological strings into "clean" agglutinating-like morphological
structures that will then be interpreted for morphosyntactic content).
I'm attaching to this e-mail a diagram I've been playing with for
Emily's and my CLS paper a few months from now. It's a more detailed
specification of the bipartite lexicon we proposed in our LSA talk.
(Actually, in the formulation in the diagram it's grown into something
like sesquibipartite since I've added the category of "lexicographic"
information as distinct from morphophonological and morphosyntactic
information.) Perhaps this will be interesting to people thinking about
the LKB's morphological support.
The morphophonological analyzer we envision has to be able to handle
just about anything a linguist working on an undocumented language
might want to throw at it--arbitrary position classes, bizarre
cooccurrence restrictions, multiple exponence of morphological
categories (like person agreement), suprasegmental morphology, etc. In
the current conception, the LKB still does some morphological analysis,
but this is limited to analyzing what are effectively agglutinating
structures, which are an attempt to abstract away from the problems of
"fusion".
Jeff
-------------- next part --------------
A non-text attachment was scrubbed...
Name: MontageMorphophon.pdf
Type: application/pdf
Size: 32986 bytes
Desc: not available
URL: <http://lists.delph-in.net/archives/developers/attachments/20050218/966df5a2/attachment.pdf>
-------------- next part --------------
More information about the developers
mailing list