Re: Consistency Checker

From: "Peter Crowther" <Peter.Crowther@networkinference.com>
Subject: RE: Consistency Checker
Date: Tue, 14 Jan 2003 17:59:54 -0000

> 
> > From: Jeremy Carroll [mailto:jjc@hplb.hpl.hp.com] 
> > - the agreement in the IRC is for five consistency checkers:
> >    Sound Lite/DL/Full
> >    Complete Lite/DL
> > But the first three are identical, so I have condensed them 
> > to one thing.
> > 
> > Have I understood correctly, or is there some subtle 
> > difference between an 
> > incomplete OWL Lite consistency checker, and an incomplete OWL Full 
> > consistency checker that I have missed?
> 
> They're not *necessarily* different, as a sound but incomplete reasoner
> merely never has to say 'yes' when it should say 'no' and therefore can
> always say 'no' --- Sean's idea of a pipe to /dev/null coming in again
> :-).
> 
> However, they may be *practically* different.  It may be easier to write
> a sound and relatively (but not fully) complete reasoner for Lite than
> for Full, for example; but it may be easy to write a sound but
> differently-incomplete reasoner for Full that doesn't do parts of Lite.
> The easy stuff at a given level doesn't necessarily match the easy stuff
> at 'easier' sublanguages.  So I'd expect to see a plethora of sound but
> incomplete consistency checkers, and (hopefully!) some sound and
> complete ones for Lite and DL.

Also, a sound OWL Lite consistency checker would not be required to even
accept documents that were not OWL Lite, whereas a sound OWL Full
consistency checker is obligated to not barf on such documents that are OWL
Full documents.  I suppose that you could build a sound OWL Full
consistency checker from a sound OWL Lite consistency checker by simply
absorbing any barfing and returning a ``don't know'' answer.  This would be
a bit silly however, as recognizing OWL Lite is harder than recognizing OWL
Full.

> Does that help, or have I done my usual water-muddying again?
> 
> 		- Peter

peter

Received on Tuesday, 14 January 2003 13:15:06 UTC