RE: Proposed response to Martin Merry, HP

On May 15, Smith, Michael K writes:
> I am happy to add "desirable computational properties" to the 
> species description of Lite in the Guide.
> But I am a little hesitant regarding "maximal decidable subset, although 
> subject to a higher worst-case complexity".  Has anyone demonstrated
> that that DL is the maximal decidable subset of Full?  

I believe that you are right to be hesitant. It would be difficult if
not impossible to prove such a statement - the expressive power of
different "subsets" may not even be comparable. I'm not even sure I
would describe DL as a subset of Full - it certainly isn't a subset in
the usual sense as it has a completely different semantics.


>The DL text
> currently says:
> > OWL DL supports those users who want the maximum expressiveness
> > without losing computational completeness (all entailments are
> > guaranteed to be computed) and decidability (all computations will
> > finish in finite time) of reasoning systems.  
> Which seems to pretty much capture the intent.  ??
> - Mike
> Michael K. Smith, Ph.D., P.E.
> EDS - Austin Innovation Centre
> 98 San Jacinto, #500
> Austin, TX  78701
> phone: +01-512-404-6683
> email:
> -----Original Message-----
> From: Jim Hendler [] 
> Sent: Wednesday, May 14, 2003 7:11 AM
> To: Jeremy Carroll;
> Subject: Re: Proposed response to Martin Merry, HP
> (note - there is a suggestion to editors for some wording changes to 
> Ref and Guide in this message - it is before the section marked 
> personal opinion)
> At 8:44 AM +0300 5/14/03, Jeremy Carroll wrote:
> >In
> >
> >
> >DanC:
> >>  On closer examination of the comment, it seems
> >>  to be more about what goes in OWL DL than
> >>  what goes in OWL Lite.
> >
> >And in ...
> >
> >DanC:
> >>Please help me find the relevant decisions
> >>and/or find evidence that those implementations
> >>pass some relevant tests and/or add an
> >>issue to the issues list.
> >
> >In January, we agreed a definition of a "complete OWL DL consistency
> checker",
> >if we had evidence that such a thing existed, and/or that more than one
> would
> >exist in the future (and the WG was satisfied that they would be
> practically
> >usable, rather than essentially theoretical exercises) then we could
> respond
> >with a message that indicated that, and that we thought that that was
> >sufficient to justify the DL level.
> >
> >If we don't have such evidence then I agree with
> >
> >
> >DanC:
> >>  Mr. Merry's point, "We're concerned that OWL users should have their
> >>  expectations met when they use OWL compliant systems." seems well
> >>  made, no?
> So let's set their expectations correctly -- we did try, by the way - 
> in Reference, section 8.2 we say:
> In particular, the OWL DL restrictions allow the maximal subset of 
> OWL Full against which current research can assure that a decidable 
> reasoning procedure can exist for an OWL reasoner.
> and in 8.3 we say
> The limitations on OWL Lite place it in a lower complexity class than 
> OWL DL. This can have a positive impact on the efficiency of complete 
> reasoners for OWL Lite.
> In fact, my original response to Mr. Merry was going to be that we 
> had already addressed his comments and point out these quotes --- 
> However, his comments and a couple of others we received show that we 
> haven't made the difference clear ENOUGH in our documents.  (For 
> example, in section 1.2 we don't mention the computational issue).  I 
> therefore suggest that editing Ref and Guide to set expectations is 
> the correct solution - consistent w/WG decisions in the past.
> One we could fix ref, is to make it clear that the difference between 
> OWL Lite and OWL DL with respect to this computational issue is 
> there.  For example, when we first introduce Lite in section 1.2 of 
> ref we say:
> OWL Lite is particularly targeted at tool builders, who want to 
> support OWL, but want to start with a relatively simple basic set of 
> language features.
> instead of saying it is known to have a relatively efficient decision 
> procedure (and citing the literature).  Maybe simply adding a 
> sentence after the one I cite above that says
> "In addition, OWL Lite is designed based to fit into a known 
> computational class that, while exponential, is lower than the 
> complexity of OWL DL [cite something]"
> I also think the "Species of OWL" section of the Guide is also less 
> clear than it could be, and might be wordsmithed to make the issue 
> clearer (for example, OWL Lite could say "Desirable computational 
> properties" and OWL DL could say "maximal decidable subset, although 
> subject to a higher worst-case complexity")
> Guus, Mike S -- would making these edits be acceptable?   If so, I 
> would include in the response to Merry and to the other similar 
> issues.
> >(A danger is that if OWL DL is tainted then the whole OWL brand is
> tainted).
> >>Jeremy
> What I would say would make OWL DL "tainted" would be to remove oneOf 
> and hasValue.  hasValue is used in about 10% of the ontologies in the 
> DAML ontology library, and oneOf, although not heavily used in that 
> library, is IMHO necessary for mapping existing sources into 
> ontologies --my group has used it in many cases where we have used 
> either an XML schema or a database schema as the basis of an 
> ontology, especially in our work with Web Service Composition [1].  I 
> would also remind the group that we actually had support in the WG to 
> put hasValue in Lite, but decided not to due to the computational 
> issue.
> I would argue strongly that it is better to explain things more 
> clearly in our documents than to change the language.  We spent a 
> long time developing a language that is well balanced for many 
> considerations, and I'd like to see if used in practice before we 
> start cutting useful features because of computational issues that 
> may rarely or never arise in real applications. For instance, 
> PARKA-DB [2], still the fastest ontology management system deployed 
> to date, is in the same complexity class as OWL DL, but somehow 
> people don't seem to mind since it can answer most useful queries in 
> a few milliseconds against ontologies with tens of thousands of 
> classes - it has a worst case time that could be in several minutes 
> for the largest ontologies built yet - but that doesn't seem to 
> matter since after 5 seconds it asks the user if they want to 
> continue, and most people say "no" and reformulate the query...
> Quoting one of our comment raisers, speaking about OWL:
> At 11:43 PM -0400 5/9/03, Bijan Parsia wrote in [3]:
> 4) Get the damn thing out the door.
> [1]
> [2]
> [3] 
> -- 
> Professor James Hendler
> Director, Semantic Web and Agent Technologies	  301-405-2696
> Maryland Information and Network Dynamics Lab.	  301-405-6707 (Fax)
> Univ of Maryland, College Park, MD 20742	  240-731-3822 (Cell)

Received on Thursday, 15 May 2003 17:04:09 UTC