Re: imports + metadata

Sandro Hawke wrote:
> I'm brainstorming here (sitting in OWLED)....
> 
> I wonder about having both:
> 
>       import   and   importMeta
> 
> 'import' of a RIF document would merge in the rules in that document

With renaming of locals presumably.

> 'importMeta' of a RIF document would merge in the metadata and also the
>       triples which encode the syntactic structure (which we haven't
>       standardized but we should, and Axel made a proposal [1]

Sounds fine for metalevel hacking but feels like a phase 2 issue.

> 'import' of an OWL XML file [2] or an RDF/XML-file which is an
>       owl:Ontology would (conceptually merge in the OWL-DL axioms,
>       ignoring all triples not playing a role in the ontology

How do you know an RDF/XML file contains an OWL Ontology, specifically 
one to be interpreted as DL? You can't. That why we original proposed 
the DataSet ontology to allow us to describe data models and entailment 
regimes so you can say "this is an RDFS source which I would like to 
interpret with full D-entailment semantics" or whatever.

The original data model identification proposal [*] still seems to me 
like a feasible approach (I know I'm biased :-)). Of course it would 
have to be updated to cope with how far RIF has diverged from the 
RDF-compatible form envisaged at the time that was written. You would 
basically have the object of the "import" directive be a set of metadata 
describing the source and entailment regime. If Harold/Michael's new 
metadata proposal wins the day then that would be expressed as a set of 
Frames though I don't know how to do bNodes in frames.

> 'importMeta' of such a document would give you the triples (ie the
>       triples which encode the syntactic structure of the ontology).
>       I'm not sure how you say you want OWL-Full inference or RDFS
>       inference or something

Surely the entailment regime is more relevant for import than for 
importMeta?

 > -- I think you "import" rules which
>       implement that inference, but the import is understood to be
>       symbolic -- you're allowed to use your own equivalent reasoner.

Given that not all the inference can be implemented as rules that seems 
like a slightly awkward overloading of import but it is a plausible 
alternative to the metadata approach.

Dave

[*] http://www.w3.org/2005/rules/wg/wiki/Arch/Data_Sets
-- 
Hewlett-Packard Limited
Registered Office: Cain Road, Bracknell, Berks RG12 1HN
Registered No: 690597 England

Received on Tuesday, 1 April 2008 21:48:37 UTC