Re: Why RDF Schema?

Thanks for the response ralph, it helps.

Let me see if I can successfully describe the situation:

1. RDF has no need for a generic schema language to describe syntax,
because everything in its domain will conform to the RDF syntax spec.

2. RDF does need a way to express how sets of properties can be
re-used and combined to describe different "classes" of resources.
This is the primary purpose of the "RDF Schema" specification.
It currently proposes inheritance for that mechanism (as opposed
to aggregation or prototyping, for example). For example, this
is so that one could express the additional or different statements
that can be made for subjects along the chain of "Snow Leopard",
"Big Cat", "Cat", "Mammal", "Animal".

3. RDF also needs a way to define the (data) types of the properties
(statements) themselves. RDF Schema has gone partway down this road with
its capacity to declare minimum, maximum, multivalue, and multiordered
properties. It also introduces "subPropertyOf".
It is these sorts of capabilities that are in direct overlap with
a generic XML schema specification (like DCD). Since RDF has weaker
requirements -- how hard is it to declare the 15 simple strings of Dublin
Core? :) -- RDF Schema should find a way to leverage that incipient
XML data typing facility, and should in fact withdraw what little it currently
specifies.

4. Because RDF is XML, a proposal such as DCD-the-next-generation may
be able to fully describe any RDF schema with as much power as any
RDF Schema mechanism. This is because DCD-NG will be capable of
expressing the RDF syntax (any of the many...), as well as the data types
of all its properties. The reason to pursue RDF Schema at all, is
because of item (2) above -- the suspicion that DCD-NG will not allow
for the expression of the compositional mechanisms that RDF wants.

----

I guess where I'm coming down in (4) is that it is a little facile
to merely state that DCD-NG is about syntactic validation, and
RDF Schema is about semantic validation, and that is that.
In practical terms, this seems to flesh out solely into the ability to
express schema inheritance.

I would be surprised if DCD-NG did not have such a capability, since
it is of general utility.

(Now I move down a level to issues of the RDF Schema proposal
itself; this should probably be in a new thread.)

Furthermore, regardless of whether a DCD-NG could express
it, IMHO RDF has not yet demonstrated that inheritance as a combinatorial mechanism
is actually valuable. RDF certainly has no need for it in the current applications
it references (PICS, Dublin Core, P3P). The toy hypothetical examples concerning
mammals are unconvincing. The standard arguments concerning
aggregation vs. inheritance in programming languages are applicable here --
not to mention the very real practical problems concerning the ease of
managing the specific schema instances. Could you imagine getting Dublin Core
and PICS to somehow share a base class?

There *is* a need for proper support for versioning and schema evolution,
but that is something that is not handled yet by RDF nor any XML schema
proposal that I've seen.

(one more thread...)

While I've got your ear, when will something be done about non-embedded properties?

-mda

Received on Wednesday, 25 November 1998 17:13:23 UTC