Re: GRDDL and OWL/XML

On May 10, 2008, at 10:12 AM, Danny Ayers wrote:

> Hi Bijan,
>
> While I disagree with quite a few of your assertions, you clearly  
> feel strongly about the conclusion (no XSLT)

That's probably not a helpful summary of my conclusion(s). I'm sorry  
that it's so complex: some of my points relate to the specific  
situation of the OWLWG and others do not. :(

> and so your argument deserve consideration. But unless I misread  
> somewhere, it seems like a key piece is missing.

I think you did misread :) Or rather, you are in a different place  
than I am. One bit of background: I've been following the HTMLWG  
rather closely (including IRC discussions and blogs, etc.). I think  
there's a very different view of how the web does and should work  
than the GRDDL community has (and that I bet that, e.g., the TAG has).

> Ok, assume for a moment that GRDDL is in place. A GRDDL-aware  
> agent, on encountering an XML document will follow its nose

BTW, please remember that I do not regard "following its nose" as  
having intrinsic value and often being the wrong thing to support.

However, in this case it is *not* following its nose, at least not  
really. Compare the semantics of owl:imports and rdfs:seeAlso *in  
agents*. owl:imports has a defined semantics...it's like a "<link  
rel="stylesheet" type="text/css...". The following is predefined as  
are the actions. Compare with a <a href="dfd">alternative rendering  
of this page</a>.

> to the namespace doc, discover the appropriate XSLT,

Actually, this isn't mandated by the spec. GRDDL agents, for example,  
could hard code translations of well known and popular transforms (as  
I believe they should).

Or they could offer the user a choice of plugins for that document  
(the way browsers do now).

> apply the XSLT and do whatever it's tasked to do with the resulting  
> RDF.
>
> In effect, to the GRDDL-aware agent the source XML *is* RDF.

That is entirely independent of the particular technical mechanism  
for conversion. In this sense, for protege4 the soruce owl/xml *is*  
RDF (as much as it is anything else).

> If you remove the XSLT from the equation, the GRDDL-aware agent  
> hasn't enough information to yield the RDF.

Only for, may I say, toy GRDDL agents.

> My question is, what are you suggesting as an alternative?

There are several options, including 1) bundling a converter, whether  
XSLT or Java or, or 2) the dynamic plugin model.

> While the XML2RDF operation is procedural, all the necessary  
> information is provided declaratively - in particular, the XSLT is  
> a mapping between two formats.

First, "declarative" is a fairly meaningless term, in these debates.  
For example, in the case of XSLT, it does *not* entail "easily  
analyzable" (write a termination checker). Compared with DCGs, it's  
hard to reverse an arbitrary XSLT (ergo, it tends to provide *one  
way* "mappings").

It's a program, plain and simple.

> In principle at least, this could provide the full specification of  
> any alternate syntax.

Which in the OWL/XML case would violate DRY as we already *have* a  
specification of any alternative syntax.

But just reflect on this for a moment: Would you want to implement a  
hand coded recursive descent parser from an *XSLT* as the spec? Or  
convert an XSLT to an ANTLR grammar?

Your claim is pure magical thinking. (Actually, what does it mean  
that the *operation* is procedural? Presumably the *operation* is a  
mathematical function which may be implemented equivalently with a  
procedural language and with a declarative language (pace things like  
time/space, but even there, there should be *some* asymtopic  
convergence).)

(Sigh. I find this very frustrating. I think you are throwing some  
buzzwords around without really reflecting critically on it. Perhaps  
we should, for the sake of argument, switch sides for a bit? Or,  
perhaps you might find it methodologically helpful to start by  
presuming what I say is true and trying to argue that case.

XSLT is very tied to its processing model. Consider how the priority  
of templates work. Consider the notion of the current node. It's a  
freaking production rules system with salience and a "jump around a  
tree" processing model! This doesn't mean it's teh suck, but it's  
*not* a specification language by any stretch of the imagination.)

> A GRDDL-aware agent has prior knowledge of how XSLT mappings can be  
> applied. Right now, aside from XSLT there aren't (as far as I'm  
> aware) any other mapping languages or tools spec'd out in such a  
> way to allow the agent to transparently interpret the source XML as  
> RDF without prior knowledge on the format - the whole point of GRDDL.

Then GRDDL is definitely inappropriate for OWL/XML (or, I'd argue,  
any W3C specced format). For ad hoc, or distributed extensibility, it  
seems ok. Browsers, however, tend to use javascript for this.

> I don't see any reason why the W3C shouldn't host XSLT  
> transformations, they don't seem that much different than, say, DTDs.

Yes, but if you had read through my lengthy post all the way through  
you would have found a link to:
	http://hsivonen.iki.fi/no-dtd/

In the particular case of OWL/XML2RDF, we have:
	Duel specifications (<http://www.w3.org/2007/OWL/wiki/ 
Mapping_to_RDF_Graphs>
	Inappropriate competing (<http://owl.cs.manchester.ac.uk/converter/ 
restful.jsp>
	A known format that should be handled by a proper parser
		(the conversion is non trivial, fwiw)
	No remotely demonstrated need
		(i.e., the cases have all been speculative)

More that I'm forgetting at the moment.

Cheers,
Bijan.

Received on Sunday, 11 May 2008 23:15:11 UTC