RE: motivation for bNodes/existentials in RDF; note for parsers

Hi Massimo,

Let me try to address your concerns here:

At 04:55 05/04/2002 +0200, Massimo Marchiori wrote:
>This thread is getting overloaded, as many concurring issues are
>overlapping (and many replies are coming... ;).
>Rather than parallelizing issues on a pile, I'll just try to focus
>on the major one.
>
>
>Let's look at the charter:
><charter>
>Implementor feedback concerning the RDF Model and Syntax Recommendation
>points to the need for a number of fixes, clarifications and improvements to
>the specification of RDF's abstract model and XML syntax. There is also
>considerable interest in the exploration of alternative XML serialization
>mechanisms for RDF data. The role of the RDF Core WG is to prepare the way
>for such work by stabilizing the core RDF specifications. The RDF Core WG is
>neither chartered to develop a new RDF syntax, nor to reformulate the RDF
>model. However, the group is expected to re-articulate the RDF model and
>syntax specification in such a way as to better facilitate future work on
>alternative XML encodings for RDF.
></charter>
>
>What this means is that the "minimum to declare victory" has to be done (to
>"prepare the way for such work",
>but not to normatively *do* all such work).
>If along this process, bugs or ambiguities or absolute-needs arise, then
>modifications can occur, but
>they must have bullet-proof justifications.

This would appear to be an assertion about your interpretation of the 
charter.  Fair enough.  Does clarifying the meaning of M&S's "anonymous 
resources" fall within your class of "bugs, ambiguities or absolute 
needs"?  i.e. do you think we had to say something about them?


>Now, back to the RDF M&S and to what it means. Forget about the "S" part,
>and focus on the "M".
>We all agree the word "model" is rather unfortunate in the M&S.
>The "M" in M&S defines essentially an *algebraic model*. Then, it hints
>on the possible meaning it could have (alas, or luckily ;), staying at
>natural language level, without formalization.
>
>So, what should be done, according to the charter, is fixing and formalizing
>the
>algebraic model (anonymous nodes first belong in this realm), and formally
>stating what the natural language semantic descriptions present in M&S say.
>Everything beyond is at risk, and ought to have very good justifications on
>why
>it has to go into this "core fix", and not just in the later "such work" the
>charter talks about.
>
>Now, the concern here is that the anonymous node --> existential variable
>is more than the "minimal fix",

Says you.  However, others disagree.

>  and is in fact at risk to belong to the
>"such work" next part.
>It might not be the case, but for public accountability, there should be
>very good reasons
>on why we have necessarily to do this.

The properties of the so called "anonymous nodes" of m&s had long been 
under discussion.  Clarification of what they meant was necessary and the 
job of the RDFCore WG.  In clarifying the specification, the WG took into 
account the various readings of M&S and also what the WG thought the best 
answer was.  After long and difficult discussion, the answer reached is the 
one described in the current model theory WD.

>Quoting from one of the replies:
>    The definitive statement in M&S is (para 41)
>
>http://lists.w3.org/Archives/Public/www-archive/2001Jun/att-0021/00-part#41
>    [[[
>    The sentence above does not give a name to that resource; it is
>anonymous,
>    so in the diagram below we represent it with an empty oval
>    ]]]
>but in fact, the reality is more complex. Anonymous nodes are not
>"first-class"
>in M&S, they only carefully appear in some places (like containers and
>reifications),
>where in fact naming such a node would just be superfluous.

Where in M&S does it state the limited roles that anonymous resources can 
play?  Various implementors, based on their understanding of M&S, have 
implemented systems (Jena, Redland, SiRPAC) which allow more general use of 
blank nodes than your interpretation would allow.

>  The "minimal
>fix" would
>be therefore just to better state what/how anon nodes are/behave in the
>algebraic model
>(because, they then disappear at the semantic level).
>Now, the anonymous node -> existential route follows three extensions that
>depart from
>the minimal fix of M&S:
>a) grant anonymous node first-class status in the RDF algebraic model
>which in turn implies that
>b) you have to deal with an algebraic model where anonymous nodes can occur
>everywhere
>which leads to the
>c) you have to define what the semantics of these "extra" (wrt M&S)
>algebraic models are.
>
>This extension chain has then to be justified as absolutely needed, in order
>to be brought
>back into the current charter and not just left to future extensions.

I'm actually rather pleased you brought this up.  I had personally been 
concerned that I had been too conservative in interpreting the charter, so 
its rather refreshing to have someone arguing we have been too liberal.

I'd like to ask you if a substantive issue lies behind your question.  Your 
argument seems to be that you feel the RDFCore WG has exceeded its charter 
in that its has extended RDF beyond your interpretation of M&S.

I've spent a lot of time arguing over the meaning of the entrails of M&S 
and come to the conclusion that its not that productive an activity.

If you have a substantive technical objection to what the WG is proposing, 
I'd like to hear it.

>Note, not arguing one can't show it's needed, just asking somebody should
>show it's
>absolutely needed.
>[Incidentally, note the current MT doesn't tackle containers and reification
>  (useful exercise: show how much this depends by the departure from M&S with

Not yet.  We had to agree what it was going to say before we could include 
it in the model theory.  Roughly speaking, as far as the model theory is 
concerned, all containers are really sequences; the difference between a 
seq and bag and an alt is that the latter two are hints to the reader how 
an application is likely to use them.  The semantics of reification is 
similarly weak; in loose terminology, statements represent statings.  Both 
of these should appear in the next iteration of the model theory.

>a) ... ;) ]
>
>Once this is clarified, we can go deeper with the other issues touched by
>the thread.
>
>Thanks,
>-M
>
>ps Sorry if this is making losing cycles to the wg. I think it's a very
>serious issue,

I tend to the view that serious issues have serious consequences.  What 
consequences do you have in mind?
[...]

>pps Brian, DanB, if you want me not to use w3c-rdfcore-wg and to use
>www-rdf-comments@w3.org instead
>raising a formal issue there, please tell me at any moment without any
>problems (I think I'm excused as
>DanC brought me in this thread, but it's also true that if this goes on
>much, it might be as an abuse
>of the mailing list).

Right.  I'd forgotten that.  I think rdf-comments is the best place for this.

Brian

Received on Friday, 5 April 2002 04:04:09 UTC