[matrix] problem applying lexeme to lexeme rule

[Emily M. Bender]

Dear Altaf,

I think the problems you have identified are actually independent
of each other.  That is, the generation failure is not because
the HCONS value is inappopriate, but because some rule (most like
not the one you included in your message) is able to apply to its
own output.

First, let me clarify some of our confusing terminology.  Rules
are cross-classified on two dimensions: inflecting v. constant
and ltol v. ltow.  A rule is inflecting if is associated with
a spelling change rule (i.e., if the rule instance has a %prefix
or %suffix line).  In this case, the rule should be instantiated
in irules.tdl.  A rule is constant is the mother and daughter have
the same form (in this case, there should be no %prefix or 
%suffix line, and the rule should be instantiated in lrules.tdl.)

On the other hand, the ltol/ltow distinction has to do with when
a word is morphologically completed and thus eligible to be a daughter
in a phrase structure (syntax) rule.  What's possibly quite confusing
is that the feature we use to track this is called INFLECTED. Signs
that are INFLECTED + are considered fully inflected (though you
can have INFLECTED + to INLFECTED + ltol rules, which add additional
optional morphological information).

Note also that if you are monotonically adding information (e.g.,
the daughter is underspecified for CASE while the mother has a 
specific CASE value), then in fact you don't need a **-change-only-lex-rule.
For the example you give below, you might want to try const-ltow-rule,
and get rid of [CASE-INFLECT no-inflect] on the daughter and your
lexical entries.  If you make sure your lexical entries are
[INFLECTED -] (by putting such a constraint on noun-lex, say) you'll
ensure that they all have to go through some lexeme-to-word-rule.

As for the maximum number of edges exhausted problem, I think
you probably have a runaway lexical rule.  See these FAQs:

http://old-depts.washington.edu/uwcl/twiki/bin/view.cgi/Main/RunawayRule2
http://old-depts.washington.edu/uwcl/twiki/bin/view.cgi/Main/CirLexRule

Please let me know if you have any further questions,
Emily

On Thu, Jul 05, 2007 at 06:18:40PM +0600, Altaf Mahmud wrote:
> hi all,
> 
> Please help me about implementing lexeme-to-lexeme rule. I have used
> "infl-***-change-only-ltol-rule" as per LAB-05, but it doesn't apply into
> parse chart. I have also seen the corresponding problem in the "matrix wiki
> page" that, if the rule doesn't add an affix, it doesn't say LKB that it is
> an inflecting rule. So, later I tried with the "***-change-only-lex-rule".
> Same output, no parses found, because the lexeme-to-lexeme rule doesn't
> apply. I also tested the "interactive unification" feature which gives me
> successful unification. For testing purpose, at later  when I used the
> "***-change-only-lex-rule", I made the 'INFLECTED' feature of the mother to
> be '+', while taking the daughter as 'INFLECTED -'. Now it can parse, but it
> can't generate because it can't give no MRS output. I am giving
> an example of a simple lexeme to lexeme rule which just change the
> CASE-INFLECT feature of the head of a proper-noun-lex:
> 
> nom-proper-noun-lex-rule := head-change-only-lex-rule &
> [ INFLECTED +,
>  SYNSEM.LOCAL.CAT.HEAD noun &
>                                              [ CASE-INFLECT nom ],
>  DTR proper-noun-lex &
>         [ INFLECTED -,
>           SYNSEM.LOCAL [ CAT.HEAD noun &
>                                                         [ CASE-INFLECT
> no-inflect ],
>                                        CONT.HOOK.INDEX.PNG.NUM sg ] ] ].
> 
> 
> Thanks. Please, help!
> 
> 
> 
> -- 
> -Altaf