[Fwd: Re: [developers] processing of lexical rules]

[Emily M. Bender]

On Fri, Feb 18, 2005 at 04:17:43PM +0000, Ann Copestake wrote:


> I had a quick look at some of Emily and Jeff's slides already and
> this all seems very sensible.  I don't see why, in principle, there
> should be a problem about supporting morphophonology in the same db
> as is used for the LKB/PET but there may be practical problems to do
> with XFST etc, of course.  It looks as though a Windows port of the
> LKB db stuff should become possible in principle soon, btw.

That's good to hear.

> I still think that in terms of integrating an external
> morphophonology system and the LKB, by far the easiest approach will
> be for the external system to instantiate a chart (i.e. the
> retokenisation option). The alternative of making it guide a partial
> derivation involves much more detailed integration of code, and
> isn't suitable for a pipelined implementation.
>
> In terms of conceptualising this as turning funny morphology into
> Italian or whatever - I'm happy with that.  But I just want to point
> out that if you don't want to conceptualise these intermediate
> entities as things that can be modelled as strings, you don't have
> to (as far as the LKB is concerned), because the idea is to support
> instantiation of the chart with feature structures directly.  The
> morphophonology as FST paradigm does, of course, tend to limit you
> somewhat, but I think it may make sense to think of the output of
> morphophonology as structured objects.

In a sense I think we're imaginging leveraging the LKB's own
morphological analyzer.  We'd use XFST to produce a tokenization
(into words) which the LKB could then easily perform morphological
analysis on since XFST had taken care of all the funny spelling
changes and left only simple concatenation.

I can see getting to structured objects of the form

[ STEM <        >
  INC-STEMS <    ,   >
  AFFIXES <        > ]

Anything more analyzed that than and I wouldn't want to try to
do it in a system that's really meant for morphophology.

Emily