Re: accuracy check on OWL-DL reasoners

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) as cases of incompleteness of an implemented system? 

Best,
-gstoil


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

> 
> On Nov 7, 2007, at 9:34 PM, Michael Kifer wrote:
> 
> >
> > It may be now, but it was not so a year ago. My info was outdated.   
> > I see
> > that Racer has announced an upcoming complete version, and Pellet has
> > become complete some 6 months ago.
> 
> This is not true. Pellet was complete almost in 1.3 beta, so sept  
> 2005. FaCT++ became complete for OWL DL some months after that. Racer  
> has had other design priorities.
> 
> Pellet and Racer over a year ago (before OWLED 2006 in Nov 2006) both  
> became complete SROIQ reasoners (i.e., OWL 1.1).
> 
> > But my point is still valid.
> 
> Well, sorta. I wish it was made with accurate facts :)
> 
> > It took
> > quite a few years
> 
> OWL went rec in Feb 2004. So, let's see, 10 months to 2005, and 9  
> months to sept, so 19 months, which is 1 year and 7 months.
> 
> This is "quite a few" years? :)
> 
> Also, there were SHOQ and SHOQ reasoners before (FaCT, DLP).
> 
> Oh, MSPASS was complete and a decision procedure long before, I'm  
> prettysure. And Hoolet was complete, but I've not tracked down  
> exactly when. I wouldn't call these serious production  
> implementations though.
> 
> > to achieve a complete implementation after the official
> > release of OWL.  Another important point is that without the OWL
> > specification there would probably be little incentive to go all  
> > the way
> > and implement those less critical aspects of OWL.
> 
> Hard to say. The main block was the lack of a goal directed decision  
> procedure for SHOIQ, which really was quite radically different that  
> the EXPTIME logics, due to the loss of the tree like model property.  
> Uli and Ian worked on it for 5 years or so. We implemented it shortly  
> after they came up with one.
> 
> However, we knew how to implement qualified cardinality restrictions,  
> and even had user requests, but didn't until we had OWL 1.1 specs we  
> were trying to validate. So, I do agree that it can help a lot. If  
> you have known procedures, it's even a bit of a no-brainer.
> 
> No need to exaggerate to make your point.
> 
> > So, if we set the bar too low for RIF then there will be no  
> > incentive to
> > work on complete implementations of important features (like equality)
> > either.
> 
> On the other hand, people haven't really taken up the guantlet of a  
> complete OWL Full reasoner. So some "reasonableness" judgement is  
> required.
> 
> Cheers,
> Bijan.
> 



--

Received on Thursday, 8 November 2007 10:18:56 UTC