Re: POWDER and OWL

Hi, Phil.

> Thanks Andrea, I am very grateful to you for expressing this. Let me see
> if I understand you, Stasinos and the whole current debate correctly
> because, as you point out, it will set the direction of the group's work.
> 
> If I understand you correctly - and that's a big "iff" then...
> 
> 1. The model we discussed in Boston, where we create a 'primitive class'
> (A) of descriptors (like mobileOK or Child Safe) and another class (B)
> that uses string matching to describe a set of resources in terms of
> their URI patterns, and thirdly create the critical relationship that B
> is a subset of A with suitable reification statements, DOES express the
> semantics we intend for a DR.

Yes.

> 2. Such data is NOT processable directly by an OWL reasoner.

Not by the available ones (at least, as far as I know).

> 3. In tools that support them, rule languages like N3 Rules and SWRL CAN
> be used to process the data. For example, one could express a rule in
> one or other of these languages that said something like "If the
> resource is available from example.org and the statement that it is a
> subset of mobileOK was made by an entity satisfying condition X and
> within a time frame Y, THEN include the hyperlink in the page." (I'd
> love to see such a rule!)

What I can say is that a DR can be expressed by using a rule language,
such as N3. Your rule is something different from a DR (actually, it
seems a rule aiming at processing a DR): which is its purpose?

> 4. Given a rule like that a reasoner could process the data, BUT, using
> SPARQL queries, one could extract the data from the DR and process it in
> a non-DL reasoning environment if required.

Correct.

> 5. The model we're discussing, whilst not processable by existing off
> the shelf reasoning tools, DOES NOT create inconsistencies. From my own
> experiments with OWL in Protege, although not a closed world in the DL
> sense, the correct inference is drawn that the resources  in B inherit
> the datatype properties of the description class A.

Correct.

> If the above is correct then my proposal to the group is:
> 
> A) We stick to this model and write it up in new drafts of our two
> primary tech documents ASAP.
> 
> B) We include examples of rules, written in N3 and SWRL, that show how
> we foresee the data being interpreted in a DL environment.
> 
> C) We further include examples of how to extract and interpret the data
> without semantic rules.
> 
> D) As discussed in Boston, we consult with the wider community, in
> particular the OWL community, and consider their advice.
> 
> This would allow us to define how to publish DRs and show clearly how to
> interpret the data WITHOUT prescribing a single 'MUST USE' method of
> working.
> 
> From my current perspective (midnight in a hotel bar in Rome), that
> sounds good!

I agree. However, as far as points B and C are concerned, I think it is
better to include them in the primer (maybe we can have a summary in the
DR doc, as in its current version). The fact that rule languages,
SPARQL, etc. can be used to process DRs is an implementation issue, so,
according to me, the DR and grouping docs are not the right places were
to discuss them.

Andrea

Received on Friday, 23 November 2007 08:56:10 UTC