[Patrick Stickler, Nokia/Finland, (+358 40) 801 9690, [log in to unmask]]


----- Original Message -----
From: "ext Tim Cole" <[log in to unmask]>
To: <[log in to unmask]>
Sent: 15 December, 2002 19:54
Subject: Re: RDF typed literals and DC encoding schemes


> ----- Original Message -----
> From: "Patrick Stickler" <[log in to unmask]>
> To: <[log in to unmask]>
> Sent: Friday, December 13, 2002 2:39 AM
> Subject: Re: RDF typed literals and DC encoding schemes
>
>
> > [Patrick Stickler, Nokia/Finland, (+358 40) 801 9690,
> [log in to unmask]]
> >
> >
> > ----- Original Message -----
> > From: "ext Thomas G. Habing" <[log in to unmask]>
> > To: <[log in to unmask]>
> > Sent: 12 December, 2002 18:00
> > Subject: Re: RDF typed literals and DC encoding schemes
> >
> > ...
> >
> > > I still think that even if XML Schema
> > > and RDF are not a perfect match, it would be useful to use XML Schema to
> > > validate at least portions of an RDF/XML document, which means that RDF
> > > should support the xsi: attributes w/o complaining and probably ignore
> them
> > > for purposes of RDF-specific processing.
> >
> > It doesn't require that. With the RDF syntax defined for typed literals,
> you
> > can (as just one possible option) run the RDF/XML instance through a
> filter
> > that maps the rdf:datatype attributes to xsi:type to validate all datatype
> > values which have definition in an XML Schema.
>
> Do you have such a filter available today?

No, but ...  sed -e 's/rdf:datatype/xsi:type/g'

> >
> > Now, that is asserting a number of presumptions that the RDF specs
> > should not themselves assert, namely that any member of rdfs:Datatype
> > also fully conforms to the XML Schema spec and has an XML Schema
> > definition (it might not and need not).
> >
> > But as an approach to validation, it's fairly straightforward.
>
> How doable is it (or will it be) to do the converse? That is, will it be
> feasible to map xsi:type values to rdfs:Datatype values for purposes of
> validation (granted that xsi:type does not support the full range of values
> and specificity  that rdfs:Datatype does/will)? This would presume that
> xsi:type values are essentially a subset of some range of rdfs:Datatype
> values. It also would presume that accepted conventions would exist for
> handling xsi:type value ambiguities (as viewed in comparison to
> rdfs:Datatype values).  A comment in one of your previous notes suggested
> that such mappings may be in the works. ("...there are plans to initiate a
> joint effort between the two WGs to define an RDF Schema which clarifies and
> explicitly states the RDF significant semantics of all predefined XML Schema
> datatypes which is either implicit or ambiguous in the current XML Schema
> specs..."). Or am I misinterpreting?

You are misinterpreting.

The only work anticipated is in clarifying the relationships between
the predefined XML Schema datatypes in terms of RDF datatyping semantics
(rdfs:subClassOf relations and the like).

Though, since RDF datatyping is intended to support all XML Schema
definable simple datatypes, including user defined types, it should
be possible to have a reasonably consistent bi-directional mapping
between rdf:datatype and xsi:type attributes on elements with lexical
form content.

> >
> > Ultimately, though, given all the other checks and tests that could/should
> > be done on RDF expressed knowledge, I expect that XML-based validation
> > will offer less and less utility, particularly when we start to see OWL
> > savvy validators, so any benefit from such an approach would be IMO short
> > lived.
> >
>
> Quite possibly. Eventually. I'm not sure I agree about the short-lived part.
>
> The problem right now of course is that we don't have commercial or
> production-quality OWL savvy validators. As best I understand we don't have
> any commercial or production quality validators that can validate against
> still draft rdfs:Datatype specs. We do have commercial and production
> quality validating parsers that understand XML Schema and can do xsi:type
> validation. Granted xsi:type is a more limited approach to datatyping, and
> granted the attraction of the RDF triples way of describing content, the
> decision by the RDF WG that to accommodate xsi:type semantics within an RDF
> context (either by ignoring xsi:type attributes or by providing an
> authoritative canonical mapping from xsi:type to rdfs:Datatype semantics
> that could be implemented by RDF parsers) would put RDF on too much of a
> slippery slope is still unfortunate, even if it is correct, as you cogently
> argue in this and recent notes.
>
> The upshot of the decision is that for many of us now generating XML
> serialized metadata in DC and related schemas RDF will remain of academic
> interest only. I was hoping that we were more near a point when we could
> begin creating metadata in RDF using an XML serialization that would still
> also be useful for immediate practical purposes (e.g., could still be
> validated, albeit at a less precise level, by existing XML Schema based
> validating parsers). Sadly, your argument suggests that this not yet the
> case, and likely never will be. Since I am constrained for the present to
> participate in environments that rely on XML Schema, and in particular
> xsi:type based validation, I will continue to forego use of RDF as I
> continue to generate new metadata records.

XML Schema validation is primarily just a check on the lexical form. I.e.
is the specified lexical form a member of the lexical space for some datatype.
It leaves then all other operations up to the applications consuming the
XML content.

Thus, if one wants to compare the equality or ordinal relationship between
two values, one is going to have to construct an application to do that anyway,
even if only working with XML Schema typed values.

The trick of mapping rdf:datatype to xsi:type is really just for lexical
checking, and only for datatypes defined with XML Schema. But any real
serious work with datatype values is going to require a proper datatyping
engine, and that will not (if done right) be tied to either XML Schema
or RDF serializations but will work with either.

> Hopefully, it will not be too costly to retrospectively convert millions of
> non-RDF metadata records to RDF at some (distant?) future date. Perhaps the
> RDF WG will provide tools to help with this process? When the time comes?

The RDF Core WG probably will not. The RDF community likely will.

Patrick