[developers] coreference problem fixed

[Emily M. Bender]

On Fri, Aug 15, 2008 at 4:34 AM, Ann Copestake
<Ann.Copestake at cl.cam.ac.uk> wrote:
>
> ebender at u.washington.edu said:
>>  When there is ambiguity in the orthographic subrules, the LKB finds multiple
>> derivations that all involve exactly the same lexical rules (and lexical
>> entries and phrase structure rules).
>
> could you be a bit more specific here?  examples would help

Sure.  Sorry for the cryptic note before.  Here's the proper bug report:

In the Wambaya grammar, I have pairs of lexical rules like the following, that
standing a feeding relationship and each have a rather messy (but, I believe,
motivated) set of orthographic subrules:

%(letter-set (!c bdgjklmnrwy))
%(letter-set (!l aiubdgjklmnrwy))

iv-gen-abs :=
%suffix (a i) (!l !l) (!c !ca) (gala galaa) (bujil bujili)
iv-gen-abs-lex-rule.

i-orig :=
%suffix (a in) (i in) (u in)
i-origin-lex-rule.

When I parse a form like "Gagarrinji" (whose
stem is "gagarra") those orthographic subrules allow two paths through
the same lexical rules:

 (a i), (i in)
 (a a), (a in)

(In either case, one more rule does (n nji) to get the final form.)

It appears that nothing in the later processing is checking whether
these two paths are equivalent at the level of the lexical rules
selected, and so I get two separate derivations.

A grammar with this behavior can be downloaded from:

http://faculty.washington.edu/ebender/wmb-morph.tgz

The string "Gagarrinji" is a suitable (short) example.

Thanks,
Emily