RE: ISSUE-66 (mapping inconsistencies): REPORTED: inconsistencies between mapping rules

Hello,

The 1.0 spec didn't have the mapping in the reverse direction, and I believe this caused quite q few problems and incompatibilities
between the tools. People had to look at nonnormative docs such as the one by Sean Bechhofer
(http://wonderweb.man.ac.uk/owl/parsing-03-10-09.shtml) to implement the spec.

The translation into the reverse direction is definitely nontrivial. Hence, to avoid mismatches between implementations, it seems
like a good idea to provide a well-defined normative translation. The existing translation does not really dictate in great detail
how things are to be implemented; however, what it does is it makes the result of the transformation unambiguous. In particular, it
addresses many of the quite tricky corner-cases related to typing and similar problems.

To summarize, I believe that having a translation in both directions should only improve the spec by making it clearer and easier to
understand. Ultimately, this should lead to fewer compatibility issues between implementations.

Finally, the existing backward translation is not really all that different from what is implemented in most OWL DL tools. We may
look at this in more detail; however, I would be surprised if we found major problems. (And if we find them, we should definitely
fix the translation.)

Regards,

	Boris

> -----Original Message-----
> From: Jeremy Carroll [mailto:jjc@hpl.hp.com]
> Sent: 21 November 2007 11:45
> To: Boris Motik
> Cc: 'OWL Working Group WG'
> Subject: Re: ISSUE-66 (mapping inconsistencies): REPORTED: inconsistencies between mapping rules
> 
> Boris Motik wrote:
> > Hello,
> >
> > There is no inconsistency here: if you serialize an OWL Functional Syntax ontology into RDF and
> read it back, you'll get the same
> > ontology, modulo pair-wise axioms (such as DifferentIndividuals).
> >
> > I don't really believe that we need a proof for that; however, I agree that saying this explicitly
> in the document is a good idea.
> >
> 
> 
> Abstractly, each set of rules defines a relationship between trees and
> triples.
> 
> 
> The design is that relationship, not the rules.
> 
> i.e. an OWL Syntax component that has a fairy inside who reads trees and
> outputs triples, and can read triples and output trees according to that
> relationship, is a legal implementation: despite the fairy acting
> entirely on magically inspiration rather than any version of the rules.
> 
> Thus having two sets of rules defines the relationship twice: and may be
> two normative descriptions of the same aspect of the design. In general
> two descriptions describe different things, even when there are people
> asserting that they describe the same thing.
> 
> I am merely highlighting that this is a potential source of problems.
> 
> Jeremy

Received on Wednesday, 21 November 2007 11:57:34 UTC