[developers] minor LKB code changes

Berthold Crysmann crysmann at dfki.de
Fri Sep 23 17:37:52 CEST 2005

Ben Waldron wrote:

> Berthold Crysmann wrote:
>> Berthold Crysmann wrote:
>>> Ann Copestake wrote:
>>>> Summary:
>>>> change to interface for `Apply all lex rules'.  info re algebra code
>>>> Details:
>>>> I have changed the interface function `Apply all lex rules' that 
>>>> can be called
>>>> from a display of a lexical entry.  Originally this produced 
>>>> separate windows
>>>> for each new feature structure - a very bad idea with a reasonable 
>>>> number of
>>>> lexical rules.
>>>> A couple of years ago, I changed it so it produced a summary of
>>>> applications in a window with the format `lex-d + rule (+ rule)* = 
>>>> new orth'.
>>>> It applied recursively and was controlled only by
>>>> *maximal-lex-rule-applications*.  But this is also a bad idea for 
>>>> grammars such
>>>> as the current ERG, which now have a rather large fanout of 
>>>> punctuation rules
>>>> given that *maximal-lex-rule-applications* is set to 7 by default 
>>>> (and I
>>>> believe that's fine for applying the morphology in the usual way).  
>>>> So the
>>>> interface now does not attempt recursive application, but the user 
>>>> can click on
>>>> a result and manually call `Apply all lex rules' again, and can 
>>>> repeat this as
>>>> often as desired.
>> Could this behaviour be toggled? Or configured by some variable? In 
>> the German grammar orthography
>> is done by a cascade of spelling rules, and I would like to be able 
>> to quickly check whether the intended form is amongst those 
>> generated... But, I can also imagine that I would not want all 
>> surface forms all the time...
> We had the same problem when working on the NorSource grammar. I added 
> a hidden parameter which toggles this behaviour.
I'll try it out!!!

> I've just checked in this code, so you can use it too. The parameter 
> is *lex-rule-show-one-step*. It defaults to T, but if you set it to 
> NIL you will see all morphological realisations. Of course, if there 
> is a black hole in your morphology component you may then get sucked 
> in...
> What did you think of my proposal regarding the format of the "Apply 
> all lex rules" list in GUI?
With multiple steps and different depth of the rule cascade, 
left-aligning sounds like a nice feature.


> - Ben

More information about the developers mailing list