[developers] processing of lexical rules

Berthold Crysmann crysmann at dfki.de
Thu Feb 10 13:35:29 CET 2005

Bernd Kiefer wrote:

>Hi all,
>i just want to describe shortly the solution i implemented in
>negotiation with Berthold without being able to go into the linguistic
>details about what is needed for the orthographemic component. Berthold
>will send a mail about his point of view soon, but maybe not today or
>We decided to re-install the application of the rule filter in the
>morphological component of PET. This give the grammar writer the
>possibility to exploit the restrictions about the successive
>application of inflectional rules in the unification part already
>during morphological analysis. Using the filter, which is optional, of
>course breaks (in general) the intermediate application of lexical
>rules (without spell changes), and might also provide a pitfall for an
>unsuspicious grammar writer.
>Berthold told me that he would be willing to sacrifice this (and have
>all lexical rules in one block before or after the spell change rules)
Just a minor clarification: lexical rules may apply both before and 
after the morphological rules, but the morphological rules curently have 
to apply in one block. We also discussed alternatives to this, e.g., 
using only features under a certain path to control the ordering of 
spelling rules, which would be used to derive a second rule filter. Such 
a move gives the grammar writer full control over the order of 
application and it will also make it possible to interleave inflection 
rules and spelling rules.

>if he would get the restrictions during morphological analysis. One of
>his main objections against an external morpology component is, that
>this makes it necessary to maintain to different lexical knowledge
>sources, whereas he would like to describe the whole thing in one
Another reason is that I would really like to have a generative approach 
to regular and productive morphology that is tighly integrated with the 
morphosyntactic side. With the exception of compounding, the current 
functionality appears sufficient to implement  both inflection  and 

>One obvious problem with this approach is that constraints come from
>two different sources: the lexicon entry of the base form and the rules
>for spelling changes. It is of course prohibitively expensive to expand
>all possible inflectional (and maybe interspersed lexical) rules
>top-down to arrive at a point where either no base entry exists or it
>is not compatible with the constructed feature structure.

More information about the developers mailing list