Re: ShEx relation to SPIN/OWL

On Aug 1, 2014, at 3:15 AM, Bernard Vatant <bernard.vatant@mondeca.com<mailto:bernard.vatant@mondeca.com>>
 wrote:

Peter

Although I agree with what you write in principle, there is an underlying practical problem, which we face daily in the management of the LOV repository.
Again, it's not a minor point that RDF uses URIs for classes and properties.

I don't understand.  Is it a good thing that RDF uses URIs or a bad thing?

If you find a class or property URI somewhere in data (e.g., DBpedia) how do you figure the semantics of this URI, IOW how do you find the document(s) you are speaking about which define it, and which ones will you abide by, to figure in practice the closure of the description of this URI?

Well, that's a good question, but that's a problem with any setup where you encounter identifiers that others have coined.  RDF doesn't make the situation any worse, as far as I can tell, *except* that it does not have an explicit mechanism (like owl:imports) for pointing to some other place to get extra information about identifiers.

Look at the result of this query  http://bit.ly/UPGA25  ran on current LOV database of 450 vocabularies/ontologies. It yields all elements related to foaf:Person in various namespaces (and so many documents). Somehow each of those 300+ assertions is refining the semantics of foaf:Person.

To some extent, yes, but one can argue that just being the domain or range of a property or the superclass of some other class is not really changing the true meaning of a class.

If you ignore the 25 ones defined by the foaf: namespace itself, you are lefy with more than twenty different namespaces/vocabularies/documents to explore, and select or not to figure the closure.

Yep, it's a problem.  However, again, this is not an RDF problem per se, just a problem with using foreign identifiers.

Of course, looking only at the FOAF vocabulary using a HTTP GET on foaf:Person, you will not access and even be aware of all this extra semantics, and you can ignore it, apart of those which FOAF itself reuses.

The suggested solution in RDF, of course, is to use the identifier as a retrieval tool and use the information so obtained.  This has lots of problems.  One of these problems is that a particular document is the one place to find out the definition

This is an extreme case maybe, but it illustrates what the practical problem will be, when using a URI, to first find out to which semantics you abide by before passing it to reasoners or constraint checkers. One could imagine something like a content negotiation on the URI where you pass in parameters the type of description you want for this URI (give me SPIN, OWL, Shapes ...). But given the messy way content negotiation is already implemented after httpRange14 resolution (based also on our LOV experience), I'm very skeptical about such a solution in practice.

My view here is very much along the lines that you determine which ontology documents you abide by, and that there can be very good reasons (messy data, resource constraints, meaning drift) to use different ontology documents for the same vocabulary at different times.

Moreover, different parties will use different documents of which the original URI owner does not control, is not aware of, let alone agrees with.

I don't see this as necessarily a bad thing.  Sure, the URI minter should be give some deference, but who is to say that a community would be better served by going with a different definition of, for example, foaf:person?

My concern is to see a door wide open to more balkanization of the Semantic Web, each tribe abiding by its own, not necessarily public, interpretation of shared vocabularies.

Well then I guess that you would be very much against the line of thought that RDF validation is dependent on usage.  Requiring different RDF documents concerning the same information is much more of a balkanization than permitting ontology variations.

peter






2014-07-31 22:55 GMT+02:00 Peter F. Patel-Schneider <pfpschneider@gmail.com<mailto:pfpschneider@gmail.com>>:
All these complaints about RDFS and OWL appear to be based on the conception that RDFS and OWL work with a single document containing everything that can ever be said about a particular vocabulary.

However, this is a misconception.  It is certainly possible in RDFS and OWL to have multiple documents that speak to the same vocabulary.  This is done, for example, when OWL ontologies are extended.  A core document talks about the core vocabulary and other documents add new vocabulary and add new information about the core vocabulary as well.

The exact same thing can (and should) be done with constraints.  You can have a document that provides the core information about a vocabulary.  You can have multiple documents that provide constraints on this vocabulary for various purposes.  When validating some information one can pick which set or sets of constraints to apply.

This division can also be done with the axioms of the ontology.  There is no need for all the axioms of an ontology to be in the same document, leading to the possibility of having the RDFS portion be in one document and the non-RDFS portion in another document.  Other divisions are also possible.

So stating that if you use RDFS and OWL you end up with a lot of baggage and lose on reusability is completely false.

Peter F. Patel-Schneider