[lkb] Generator chart has edges for bare affixes

[Ben Waldron]

We've experienced the same problem in a grammar developed by a student 
in Trondheim. The problem disappeared when 'spelling-change-rule-p' was 
redefined to correctly define all relevant inflectional rules as 
spelling-change-rules -- in this case by defining the function to state 
that 'lex-item' and it's subtypes (rather than 'word' and it's subtypes) 
are spelling-change-rules:

(defun spelling-change-rule-p (rule)
  (let ((rule-type (type-of-fs (tdfs-indef (rule-full-fs rule)))))
    (when (or (eql rule-type 'lex-item) ;; <--
              (subtype-p rule-type 'lex-item)) ;; <--
      (let* ((mother (tdfs-indef (rule-full-fs rule)))
             (morth (existing-dag-at-end-of mother *orth-path*))
             (daughter (existing-dag-at-end-of mother '(ARGS FIRST)))
             (dorth (existing-dag-at-end-of daughter *orth-path*)))
        (not (eq morth dorth))))))

A question: Would using such a general type in the function above be 
expected to lead to any problems elsewhere?

 -Ben

Emily M. Bender wrote:
> Hi,
>
> One of my students has created a grammar with an interesting
> pathological property: when one uses it to generate, the chart
> contains edges for affixes corresponding to many of the lexical
> rules (as if they had attached to a null stem).  Any idea
> what might be causing this?
>
>
> The grammar can be found here, if anyone wants to play with
> it:
>
> http://courses.washington.edu/ling567/arabic.tgz
>
> A test case is:
>
>  ?-i?dar  ?-akal    ?al-kizaz-a
>  1SG-can  1SG-eat   the-glass-ACC
>  'I can eat glass'
>
> Emily
>
>
>