Re: FORTH COMMENTS ON RDF Schema: Last Call

Hi

* Vassilis Christophides <christop@ics.forth.gr> [2003-03-21 16:11+0200]
> Dan
> 
> I have argue several times about the RDF Schema issues presented in
> summary in my mail.

Yes, as Pat noted these are topics that have also taken up a fair 
amount of time in the RDF Core and WebOnt WGs, as well as the DAML+OIL
work that preceded WebOnt, and in the RDF Interest Group. 
 
> No I am not satisfied by Pat Hayes answer. My problem is that I don't
> have enough time to debate with excaustive details. 

I appreciate the time problem. At this stage in the RDF Core work, we would
however need to assemble a pretty compelling case that the current design
is broken. I believe everyone agrees that there are other ways that RDF 
could have been designed, and that there are costs and benefits associated
with each design. At this stage I am focussed solely on making sure 
that critiques of RDF Schema presented during Last Call get 
acknowledged and handled by the working group. I think the best thing 
to do is open a Last Call RDFCore WG issue and have the working group 
consider your concerns, as expressed here, in the original message, 
and in the papers you cite.

I am trying to break your comments down into issues we could 
productively discuss.

Two of your key concerns are:

 -rdf/s layering too flat
 -property instances aren't explicitly represented

You also present an alternate design for domain and range, a design
which has been considered at length and ultimately rejected by 
the RDF Core (and WebOnt) working groups. Similarly, you express a 
preference for the older (and previously rejected by both WGs) design 
that banned rdfs:subClassOf cycles. I could request issues be opened for
these, but in the absence of compelling new evidence, I would not expect
the WG to seriously consider changing its design.

I won't deal here with your comments on datatyping; someone else will
(or has) respond to that.

I propose we focus on your two concerns that have not been so 
explicitly discussed by the WG: layering and property instances. Shall we
request the WG open issues for the 'layering is too flat' and 'property 
instantiation' concerns that you raise?


> The axis of critic
> are:
> 
> 1) Some of the RDF/S features are not justified by real application
>    needs (at least the 2-3 applications I am dealing with)

Other's experience seems to differ. We need to finalise on a single design
so at this stage, if you are asking for a major redesign, we need evidence
of major problems.

In my experience, rdfs:subClassOf, and the associated simple 
extensional semantics associated with it, is very well suited to mixed-
namespace, open world Web applications. In a more closed environment I 
share your preference for avoiding subclass loops. RDFS, however is a 
language designed for deployment in the Web.

> 2) They imply serious implementation limitations
> 
> For example, class cycles is a very confusing notion in P2P e-learning
> applications, developers can't skip off this costly reasoning since it
> is embedded in the default RDF/S Semantics and there no efficient
> algorithms to implement transitive closure computations.
> 
> More than an official response, I need a fruitful scientific
> discussion on these issues. We at least trying to explain our
> objections in all our papers
> (http://139.91.183.30:9090/RDF/publications/FuncBook.pdf 
> http://139.91.183.30:9090/RDF/publications/comp-networks2003.pdf)

I want to make sure the Working Group are aware of your concerns, 
review comments, and alternate designs. At this stage the best way to do that
is to focus on specific issues that, given the stage we're at (post Last Call)
will have most chance of improving the RDF specifications. Revisiting 
subClassOf and domain/range discussions of 1998-2002 is unlikely to 
be productive unless new evidence is brought to bear. All viewpoints on this
(including logical, application-oriented, tool developer etc) have been 
thoroughly aired, on www-rdf-comments, www-rdf-interest, www-rdf-logic, 
and elsewhere. The working group membership have studied much of this 
material, participated in those debates, and the result was the design
published as our Last Call specs. 

I'm concerned that we take time now to distinguish any new points you're 
raising from design discussions which (in my opinion) have been exhausted
and full examined. In that light, I'm proposing that the WG open issues
for the 'flat layering' and 'property instances' concerns you raise above.

I hope my comments don't appear dismissive. My understanding is that 
your views were taken into account during the discussions of the last
few years, even if the final specifications did not match the design 
you would have preferred.

Let me know whether you would like me to proceed with the two issues above.

Thanks again for your review,

cheers,

Dan

Received on Friday, 21 March 2003 09:47:51 UTC