Re: nominals (was RE: Proposed response to Martin Merry, HP)

On May 12, Jeremy Carroll writes:
> 
> Ian:
> >>  I think this is covered by Jim's proposed response. Users should be
> >>  aware that nominals are a very powerful construct, and the
> >>  (gratuitous) use of nominals is likely to adversely affect
> >>  performance.
> Jim:
> >Jeremy - if you say something about this in Test, it could be a 
> >useful thing.  If you decide to, can you let me know so I can add to 
> >the appropriate responses.
> 
> I don't think that this sort of comment fits in test.
> It would be better in guide perhaps or reference.
> 
> Concerning test:
> 
> We might find, after we have not had successful implementation reports, that 
> tests involving both inverseOf and oneOf need to be moved to the extra credit 
> section (or some other device to mark them as too difficult); in which case 
> some explanation would need to be offered - but I would hope that would be 
> reflected elsewhere.
> 
> Concerning the comment:
> 
> A different move would be to decide that documents involving both constructs 
> are OWL Full, which would be a small substantive change, which might better 
> reflect this conversation.

One problem is that things aren't so simple. The nasty interaction
between inverse and oneOf is just one manifestation of the complexity
of OWL DL. Counting quantifiers also play a part - eliminating
counting would (I believe) make it possible to have a "practical"
reasoner for both inverse and oneOf.

Another point. In OIL, the problems caused by oneOf were finessed
using the well known technique of giving nominals a weaker semantics,
i.e., treating them as primitive classes (disjoint in the case that
the individuals are asserted to be different). This works fine for
most realistic use cases: if you look at what users are actually doing
with nominals, in most cases the choice of a nominal is arbitrary and
it could just as easily be a class. Sadly, this semantics for nominals
did not make its way into DAML+OIL or OWL. I believe that, in
practice, most OWL DL implementations will silently use this
technique, and that few users will notice. Of course I realise that
this isn't a completely (no pun intended) satisfactory solution.

Ian



> 
> Jeremy
> 
> 
> 
> 
> 
> 
>

Received on Wednesday, 14 May 2003 17:56:40 UTC