Re: rdfs:isDefinedBy (Was Re: Representing DCMI semantics asRDFschemas versus Web pages)

> So what's the use of an RDF schema?
> How can I prevent using a property in an improper way?
> e.g. I want that it's possible to say that a thing is an Author only if that thing is a Human. I don't want to say that, since A is an Author, A must be a human! That's not _validating_ (i.e. checking), that's inferring (i.e. saying new things)!

Exactly so - it is doing inference. RDF schema is a lightweight ontology
language, it allows you to derive additional entailments but does not, of
itself, express constraints (has no form of negation in it).

There are two useful ways to express constraints at the moment - use a more
expressive language or add a processing model. 

For the more expressive language the only practical game in town at the moment
is DAML+OIL - this is (more or less) an RDFS superset and does allow you to
express constraints. There are DAML tools around but whether any of them are
sufficient unto your task I can't guarantee. In the longer term WebONT will
presumably become the blessed and standardized successor to DAML+OIL.

The alternative is to add a processing model which is specific to your
application. It is fairly straightforward to build a tool that is able to do
useful validation by employing additional interpretation constraints. For
example, it might additionally assume that your schema describes a type lattice
and that all instances must be well typed in the sense of being assignable a
single unique type with the lattice (this need not exclude having additional
types in other namespaces as well). Such a tool would then be able say that if A
has type "Computer" then it can't be an "Author" if both these terms are
declared in your vocabulary but "Computer" is not a superclass of Author in your
schema. 

The point is that the additional assumptions and constraints embedded in such a
tool are going beyond the specs. Though that doesn't stop them being useful and
appropriate for a given class of applications. 

This approach is not strong enough for all applications. For example, even with
additional processing constraints, RDFS does not have enough machinery to allow
you to distinguish between properties which are required on a given class and
properties which are acceptable but optional. 

> So what type of constraint can be _validated_, according to w3c?

I'm in no way an authority on truth-according-to-w3c :-) 

As Patrick said earlier in this thread - you can validate syntax and you'll soon
be able to validate primitive datatypes. I assume the W3C answer on validating
class structures is "wait for WebONT". I think it would be possible to agree on
a useful set of processing models that would allow some particular schema-like
validation based on RDFS + additional interpretation constraints that would suit
an interesting range of applications but I'm not aware of any W3C groups working
on creating such a standardized model.

At an earlier stage WebONT seemed to be developing a layered architecture that
included a frame-like lightweight subset which would have lent itself very
nicely to schema-like validation. As I understand it that was dropped in favour
of sticking closer to DAML - great shame if true. Anyone from WebONT want to
shed more light on any of this?

Dave

Received on Monday, 10 June 2002 11:56:05 UTC