Re: Want GRDDL intro for WG-dummies :)

On 9 May 2008, at 10:06, Rinke Hoekstra wrote:
[snip]
> Yes, GRDDL does not require you to use XSLT, it can be a web  
> service as well (cf. Bijan's remark on the telecon).

Actually, it can just be an abstract function. We need to be clear on  
the difference between specs and implementation here.

Nothing about GRDDL is required for us to accurately spec the  
transformation, which is, after all, our core job here! In fact, it  
needs a bit more work to be quite good.

> But to answer the gist of your question: in principle yes, because  
> GRDDL allows you to specify multiple transformations that each  
> extract different information from the XML. It's about 'gleaning'  
> not translation. I.e. given a particular XML file, give me the  
> particular RDF triples encoded in this XML that I can understand.

Yes, so it doesn't even spec an equivalence relation, necessarily.

The other reason to "do" anything wrt to GRDDL, afaict, is so that  
authors who want to use GRDDL or readers who receive an OWL/XML  
document with a GRDDL attribute can detect whether the transform is  
intended to produce the equivalent RDF/XML document (as opposed to  
just the classes, for example).

Us supplying XSLT is the working group providing an *implementation*  
of this transform. This is what I object to.

I actually think that GRDDL isn't really the right thing for the  
equivalence transform in general. Current best practice seems to be  
namespace sniffing, which is surely adequate in this case. However,  
if GRDDL tools would like us to spec a marker for the equivalence  
relation so they can special case it and users would like the marker  
speced so they don't have to make up one, I'm ok with that. That's  
what I thought we were after!
[snip]
> Because that's not what we're after... or maybe: that's exactly the  
> issue we have been debating. A requirement most (if not all) WG  
> members share is that *if* we have an OWL/XML format it should be  
> translateable to RDF/XML (or RDF) in a standard way.

But this is already accomplished by the current specs.

> We need some way to point to this official transformation:

Well, I usually point at the specs :) Seriously, if there's a  
question about whether a program correctly implements the  
transformation, we definitely SHOULD NOT point to a particular  
implementation (whether in XSLT or Java or provided as a web service).

> this is what GRDDL standardises, and allows us to do. 'Just' a  
> pointer to an XSLT does not do the same thing.

And to answer Micheal, my objection is the inverse of what you  
supposed. I'm fine with specing a GRDDL marker (though I think it  
should be optional). I object to our providing and blessing an  
implementation of the transformation function (and I don't think  
marking it informative is sufficiently shielding; I don't think a WG  
should be in the business of providing implementations at all.)

Cheers,
Bijan.

Received on Friday, 9 May 2008 09:24:37 UTC