Re: GRDDL and OWL/XML

Bijan (cc'ing Alan over this issue),

    After waiting a few days, it appears this discussion has come to 
some sort of end - so I'll summarize what I think the message to the 
OWL2 WG over this debate is from the remaining active members of the 
GRDDL WG on this list.

    In effect, while it appears active GRDDL WG members on the list  
recognize the dangers posed by someone thinking an conformant GRDDL 
transformation in XSLT for OWL2/XML to RDF is more "normative" than some 
other transformation, it appears to be the feeling of the majority of 
the people from the GRDDL Working Group that the benefits of having such 
a GRDDL transformation accessible from the namespace document of OWL2's 
XML syntax outweigh the costs.

This is in particular  *if* the OWL2 Working Group believes that 
RDF-aware agents may want to access information in OWL2 in RDF and may 
not have their own locally installed OWL2 XML to RDF transformer.

So:

    1) It seems general opinion in the GRDDL WG that having a GRDDL 
Transformation Property point to a non-executable list of 
implementations is not a useful use of GRDDL, and the GRDDL WG did not 
explicitly discuss this possibility when creating the GRDDL 
Specification or test-cases. A GRDDL transformation, while not 
necessarily XSLT, is usually executable, and currently all examples are 
in XSLT.

    2) However, while the "cost" of alienating developers of other 
non-GRDDL transformations may exist, this is a misunderstanding about 
the "normative" nature of GRDDL transformations. In particular, a list 
of other conformant implementations could be accessible at the namespace 
document. Also, a few sentences explaining GRDDL and the fact that it is 
no more "normative" than other implementations could be included in the 
namespace document. If this does actually cause problems, these problems 
could be dealt with re the usual feedback channels of the W3C. The 
advantage of having a GRDDL from OWL 2 to RDF is that it is to a larger 
audience (RDF users without an explicit OWL 2 to RDF transform), who 
might otherwise be unable to have what benefits OWL 2. in RDF, can have 
these benefits with a minimal of work.

    3) Re the possibility of the OWL namespace document being overloaded 
by requests for the transform, we do in the specification encourage 
caching of GRDDL transformations. Regards the possibility of GRDDL 
transformations being "automatically" run, this is a matter of local 
policy. Current libraries such as Redland and Jena do allow GRDDL to be 
turned on explicitly by the local client.

    Although this is not the solution that Bijan wants per se, think 
this is the general opinion of the GRDDL WG members who have responded 
so far on this list, except for possibly Jeremy Carroll. Thanks for the 
provocative and intelligent critiques and comments re GRDDL and OWL 2.

    The choice of whether or not to include a GRDDL transformation that 
is an XSLT from OWL2 XML to RDF is, of course, in the competent hands of 
the OWL 2 WG. However, I do hope this feedback helps clarify things.

          -harry




   


Bijan Parsia wrote:
>
> On May 13, 2008, at 5:09 PM, Jeremy Carroll wrote:
>> Dan Connolly wrote:
>>
>>>> People are reading the SHOULD as a MUST.
>>> Really? I am not. I just haven't seen a good reason to make
>>> an exception in this case.
>>
>> +1
>>
>> (the definition of SHOULD is very good in my view, see RFC 2119, 
>> which I am sure Bijan knows ...)
>
> Sure.
>
> It would help me if I could see a hypothetical circumstances what you 
> think the case is successfully made to go against the SHOULD.
>
> In the OWL/XML case we have:
>     1) an existing specification (at the W3C) for a format closely 
> identified with semweb
>     2) widely available (in multiple ways) implementations
>     3) a culture of manually invoked translation (e.g., between Turtle 
> versions and OWL/RDF, etc.)
>     4) no clear use case for other sorts of user programs wherein they 
> would have grddl but not be able to manage other ways of getting 
> translation
>         And no clamor from users of those tools who are also OWL 
> users. Alan Ruttenberg *is* an OWL user, but my understanding is that 
> he's concerned more with legacy OWL oriented tools that only consume 
> RDF/XML.
>     5) a somewhat complex transformation
>
> (And let us presume there is an existing, seemingly decent, XSLT to 
> hand. Or two!)
>
> I think without a thumb on the scale (i.e., that we have a default to 
> override via a SHOULD) that the case is clear. (Do we agree?) With the 
> thumb on the scale, I also think it's clear, but evidently you and Dan 
> don't agree. (i.e., there are problems, and no clear upsides, and the 
> downsides of not doing it are easily mitigated.)  What other facts (or 
> sorts of fact) would tip the scale for you? If there are none, or they 
> are outlandish (e.g., space aliens say we shouldn't), then you and dan 
> are treat this SHOULD as a  de facto MUST. Or, at least, I cannot 
> discern the actual difference.
>
> If it's a de facto should, you should make it an actual should. It's 
> really misleading and confusing otherwise.
>
> Cheers,
> Bijan.
>
>

Received on Friday, 16 May 2008 16:08:47 UTC