[Fwd: Re: [developers] processing of lexical rules]
Emily M. Bender
ebender at u.washington.edu
Mon Feb 14 03:40:48 CET 2005
On Sat, Feb 12, 2005 at 05:22:19PM +0000, Ann Copestake wrote:
> I'm now sure that we need to go to an architecture where there are
> just charts and feature structures. Preprocessing is defined to
> return a chart instantiated with feature structures (expressed in
> XML for the external preprocessors). It can be coupled with
> morphological analysis in the case where we have external
> morphological analysers. In this case it should be possible to
> support a word syntax approach or an approach where the affixes
> correspond to lexical rules.
I think I don't have the right picture of an "external preprocessor".
The XFST stuff that I've been playing with happily returns a string,
and I can imagine adapting that to some kind of XML (though I don't
see the benefit), but that formalism is not up to feature structures.
The morphological features encoded in the string include things like
+Nom (which you could imagine changing to [ ... HEAD.CASE nom ], with
some minor effort) as well as +Causative, which just won't work that
way. Are there other kinds of morphological analyzers out there
which return sufficiently rich output, or are you imagining something
more like a word syntax approach where each morpheme gets looked
up in a lexicon and the feature structures in the chart correspond one
to one to morphemes?
>From what I've looked at so far, word syntax becomes crucial in cases
of open-class stem incorporation, but otherwise ranges from marginally
awkward (requiring, say, a rule for attaching case affixes, which,
best case [Finnish] applies to 17 different affixes) to insufficient:
the LFG folks do word syntax, but they still need a lexeme-to-lexeme
mapping to handle passive or morphological causative, in addition to
the word syntax rule which attaches the passive/causative morpheme.
More information about the developers