RE: Sketch: reasoning conformance levels

On November 8, Jeremy Carroll writes:
> 
> I wrote:
> > Reasoning components MAY claim "most of OWL DL reasoning" if
> > they provide at
> > least OWL Lite reasoning and ... [tbd] (e.g. pass 90% of the tests).
> 
> I heard:
> 90% is a very bad idea.

No you didn't - you heard that any notion of conformance based on some
inevitably subjective measure of "most" is a VERY BAD idea.

> 
> I didn't hear alternatives for the [tbd] ...

For obvious reasons given the above.

> 
> Suggestions please.

Don't tempt me.

> 
> Ian:
> >What now seems to be suggested is that the language is more or less
> >the same as OWL DL, and that ease of implementation is instead due to
> >different requirements w.r.t. implementation. Why bother with OWL Lite
> >at all in this case?
> 
> The proposal with the OWL Lite conformance is that it respects the syntactic
> subset issues, *and* OWL Lite reasoners may be less powerful than OWL DL
> reasoners, even over that syntactic subset.

You are missing the point. If we were to go down this route (heaven
forbid), there would be no reason to have a syntactic subset - you
could just weaken the conformance criteria for the OWL DL language
until ease of implementation was judged to be the same, e.g., we could
decide that 80% correct for OWL DL is just as easy to implement as 90%
correct for OWL Lite. This would save all the trouble of specifying 2
different languages.

And just in case there is any doubt, I don't intend the above as a
serious suggestion!

> However, I take it for granted that we would want an OWL DL reasoner to
> automatically be an OWL Lite reasoner, so the underlying semantics is the
> same.
> 
> I realize this is coming from a different direction than a theoretical study
> of the complexity problems of different constructs for complete reasoners.
> 
> So proposal is OWL Lite = OWL DL - some of the
> hard-to-implement-complete-reasoning-for-cosntructs
> *and* OWL Lite reasoners may be incomplete in some well defined way over
> that subset.
> 
> This leaves OWL DL as a distinguished conformance point where reasoning is
> complete.
> We may want to have a conformance description for complete OWL Lite
> reasoners, but, on your analysis it doesn't seem we would have many takers
> other than those who are really shooting at OWL DL reasoning.

I strongly disagree. I believe that the current specification of OWL
Lite is very amenable to complete implementation and that this will be
the target for many implementors.

Ian

> 
> Jeremy
> 
> 
> 
> 
> 

Received on Friday, 8 November 2002 10:04:36 UTC