Re: RIF RDF and OWL compatibility - mini review


Thanks for the comments.  Find replies in line.

> Since I won't be at the F2F here are some comments on:
> ** RDF Compatibility
> The RDF compatibility section reads well, I like the early text 
> motivating and illustrating the approach.
> Minor comment: in the examples it might be preferable to use a CURIE 
> prefix (or a full URI) for things like "uncleOf" or if this is intended 
> to be a relative URI then perhaps a '#' prefix.

agreed, I will change this.

> In the definition of "conforming datatype map" the reference to RIF-BLD 
> at the end of condition 3 should probably now be RIF-FLD since that's 
> where rif:text is now defined.

will change this.

> ** OWL Compatibility
> Minor comment: The last paragraph of "OWL Species" claims that OWL DL 
> allows punning in the abstract syntax and that this "paves the way" 
> towards OWL 1.1. Perhaps true though the punning permitted in the OWL 
> 1.0 abstract syntax is limited relative to the OWL 1.1 proposals (does 
> not include property punning).

noted. I will make a note.

> I don't like the big difference in approach between the OWL DL and 
> RDF/OWL FULL combinations.

I don't like it either, but there is simply a very big difference 
between the OWL DL and Full semantics.
In all attempts we made it to define a single semantics for the 
combinations with both species, either there would be unacceptable 
limitations in the rules about RDFS/OWL Full ontologies or the semantics 
of OWL DL would be destroyed.  See [1] for a discussion of the different 
possible combinations we tried.

> Why is the OWL DL case mapped to unary and 
> binary predicates? Why not frame formulae?

this is explained in the second paragraph of section 4.3.2.

Basically, if we want to preserve the semantics of the OWL DL ontology 
in the combination, we cannot map unary and binary predicates (classes 
and properties) to terms.

> I don't see how the fix up of constraining the truth valuation so that 
> the frame and non-frame representations are equivalent works. In BLD 
> predicates and constants are required to be disjoint so I don't see how 
> the constraints like:
>   It-dl( t [ rdf:type -> A] ) = IR(A)(t)
> can be applied.

These are not constraints, but definitions.  Effectively, they make 
frames just another way of writing unary and binary predicates.

> Either this is a pure syntactic preprocessing step (in which case case 
> define it as such, instead of via semantic fix up) or you are lifting 
> that restriction in the BLD semantics.

It could have been defined as a syntactic preprocessing step (rewriting 
frames to unary and binary predicates); this is explained just above 
section  However, I find the current definition more elegant, 
because it provides a direct semantics, rather than requiring two steps 
(rewriting and interpretation).

> In that case why not use that for 
> RDF and OWL FULL as well?

It would disallow quantification over classes and properties, i.e., no 
variables are allowed to occur in class or property positions. I believe 
that such a restriction is unacceptable for RDF-rules.  Plus, it is 
unclear to me what the consequences would be of using equality in the 
rules in such a setting.

That said, we did consider this possibility in [1] in the section 
"Semantics> OWL Full>Alternative definition of RDF compatibility for 
increased homogeneity with direct OWL DL semantics", but dismissed it 
because of the reasons I mentioned above.

best, Jos

> I realize there may not be any good answers here but this radical 
> difference is troubling on first reading.
> Dave


Jos de Bruijn,
One man that has a mind and knows it can
always beat ten men who haven't and don't.
   -- George Bernard Shaw

Received on Monday, 18 February 2008 16:08:47 UTC