Re: accuracy check on OWL-DL reasoners

Obviously, since there indeed exists a sound and complete OWL DL algorithm,
when one implements it (obviously) in a non-utterly ridiculous way he indeed
has a fairly complete system. But, this does not mean he has a complete OWL DL
implementation, because a counterexample might be found and it might turn out
that he did not implement some rule correctly. 

Anyway, if you are interested in counterexamples then maybe we can take this
off list.

BTW, neither Pellet nor FaCT++ support differentFrom and sameAs, do they?

Cheers,
-gstoil


Bijan Parsia <bparsia@cs.man.ac.uk> said:

> 
> On 8 Nov 2007, at 10:16, <gstoil@image.ece.ntua.gr> wrote:
> 
> > Hi Bijan and all,
> >
> > Do you also count "bugs" (in the sense of incorrect or inaccurate
> > implementation of the theoretical algorithm rather than exceptions  
> > and related
> > stuff)
> 
> I don't know what distinction you are trying to raise. If you  
> implement nominals using the disjoint classes approximation, then I  
> wouldn't call you complete. If you implement the SHOIN tableau with  
> the NN Rule, then, assuming it isn't utterly ridiculous, I'd count it  
> as complete.
> 
> I would also count it as complete if it implemented a (reasonable)  
> hard coded resource bound. Say, 1 year and 100 terabytes of memory  
> (or even something tied to pointer size on a particular architecture).
> 
> > as cases of incompleteness of an implemented system?
> 
> Every software system has bugs. There is not a bright light when a  
> buggy implementation ceases to be an implementation.
> 
> Similarly, naive implementations that are clearly hopeless in all  
> cases probably shouldn't "count" (at least for some purposes).
> 
> However, if Pellet counts as complete as of 6 months ago, then I  
> would easily class it as complete as of at least a year and a half  
> and more likely of two years ago. It makes a good faith effort to  
> implement correctly and in a reasonably optimized way a decision  
> procedure for SHOIN and for SROIQ.
> 
> How this bears on your decisions about whether to include a feature  
> in whatever part of the RIF is under discussion is outside my ken.
> 
> Cheers,
> Bijan.
> 



--

Received on Thursday, 8 November 2007 11:58:39 UTC