- From: Sandro Hawke <sandro@w3.org>
- Date: Wed, 25 Jun 2003 11:47:47 -0400
- To: www-archive@w3.org
[This is a transcript of an IRC conversation on a W3C internal channel, which we decided really ought to be public.] <sandro> Hrm. There's text in RDF Semantics talking about rdf:first/etc which is absolutely incompatible with semantics-are-defined-by-URI, I think. What to do about that...... <tim-cape_> incompatible? <sandro> yeah. "RDF Semantics" says for instance that { _:x rdf:first _:y, _:z } does NOT entail that _:y and _:z denote the same thing. Fine. We knew that. But then it says that "Semantic extensions MAY" do this. That is, using the *same* URIs, and RDF graph would mean two different things, depending on which "semantic extensions" (like RDFS and OWL) were being used. <sandro> Thus RDF graphs need to be transmitted in an envelope which says which semantic extensions are being committed to. <sandro> (and which ontologies are being committed to, meanwhile) <sandro> ... and dereferencing the URI is not enough. <sandro> I'm not saying we have to change the specs or stop RDF Core, but I think we really need to be honest about where RDF fits into the semantic web. And the answer, increasingly, is that you can't just transmit RDF. You need private agreements about it, etc. <tim-cape_> no no no no no <danbri> how do you feel about that, tim? <tim-cape_> the whole point about it is atht eth web protocols ARE private (rather public really) agreements. <danbri> the old M+S RDF spec didn't assert the transistitv of rdfs:subClassOf, but didn't outright dispute it, so RDFS could come along and claim it was transitive (in prose); and OWL could say that for machines. When we say that RDFCore's semantics don't (thmselves) claimg rdf:first to be functional (or whatever) I hope it is in the same vein, ie. they leave the window open for later claims. <danbri> feel free to copy that to #rdfig if context arises where it would make sense <DanC> not private agreements; the ontologies are public; the envelope is the web. <sandro> I'm having trouble understanding what Tim means by "the whole point about it is atht eth web protocols ARE private (rather public really) agreements." <sandro> I'd say: the whole point is that the web protocols are public agreements, also known as "standards". <danbri> are individual ontologies 'standards'? <sandro> Yes. Microstandards. :-) <DanC> tim's just saying that where sandro said "RDF relies on private agreements" timbl agreed, and said that web protocols serve as "private agreements" though they're actually quite widely disseminated, for the most part <DanC> individual ontologies can also form part of the agreement, though we don't expect everybody to be party to them; only those that have read them; e.g. those that use URIs from them. <sandro> Hrm. My point is that there's a gap between web protocols and RDF, where no one gets to say which extensions & ontologies they are using with their RDF. <DanC> I think the "follow your nose" principle fills the gap. <DanC> i.e. if you use ont:term, you're party to ont <sandro> Right, but Pat's prose in "RDF Semantics" actively refutes that, saying that RDF semantic extensions may define rdf:first differently. <sandro> s/may/MAY/ <DanC> not inconsistently; they can further refine it <DanC> just like rdf:type is further refined by the definition of every class <DanC> formally, in the general case, any ontology can restrict the satisfying interpretations arbitrarily. <TimB-L> Sandro, people should not be disputing what rdf:first means - they should be disputing simply which set of inference axioms they are going to expect of a given operation, quesry etc. * sandro considers Dan's argument.... <DanC> s/not inconsistently/not non-monotonically/ <TimB-L> I ahve just printed off http://www.w3.org/TR/rdf-mt/ <sandro> I don't see the difference there, Tim. Can a collection have more than one first element? Doesn't the answer to that affect what {x rdf:first y, z} means? <DanC> er... that dates from Feb, no tim? <TimB-L> You are saying that the problem only exists in a later version and you can't get ca consensus on #rdfig about which version it is. * DanC isn't sure what problem tim is looking for <DanC> oh.. the first/rest thing... dunno if that's changed much lately <sandro> I'm saying the problem exists in that version -- I haven't looked at the later version, but I thought I/we should. <TimB-L> me l;ooking for problem: Sandro: Hrm. There's text in RDF Semantics talking about rdf:first/etc which is absolutely incompatible with semantics-are-defined-by-URI, I think. What to do about that...... <sandro> But DanC has a different angle, I suppose. That Pat's saying while rdf:first means one thing, people are free to additionally constrain it. <TimB-L> No, not nayone is free to additionally constrain it. <danbri> but not contradict it... <TimB-L> But the owners can explain more stuff about itas they find out how. <danbri> <sandro> yeah. "RDF Semantics" says for instance that { _:x rdf:first _:y, _:z } does NOT entail that _:y and _:z denote the same thing. <DanC> I thought we agreed that OWL could further constrain rdfs:range, timbl <danbri> 'does NOT entail' is different to 'doesn't say that it entails' <danbri> isn't it? <TimB-L> Might not be. <TimB-L> entail under _______ <danbri> yup, rdfc-entail... <TimB-L> "entail under the axioms gioven in this document" <TimB-L> So there are different sets of axioms one can use, all subsets of some imaginary set of all true axioms written in some imaginary suitablly expressing language. <sandro> (Pat actually does use the words "does not entail" for that example. 6th paragraph of http://www.w3.org/2001/sw/RDFCore/TR/WD-rdf-mt-20030117/#collections Of course he means "doest not [RDF simple] entail" <TimB-L> He ios talking about the existence of lists there <TimB-L> he has <ex:aaa> which may not be what he means. Weird URL. <TimB-L> I think he means ex:aaa <DanC> it's a wierd URL alright, but I asked, and he really does mean <ex:aaa> <TimB-L> { ( ex:aaa ex:bbb ) } => { ( ex:bbb ex:aaa ) }. <TimB-L> We should him to stop encouraging people to invent new URI schemes. <sandro> axioms are kind of a middle ground between formal semantics and running code. "RDF Semantics" says there is "RDF simple entailment" and then there are extensions. The rdf:first example entailment I gave is a non-entailment under "rdf simple entailment" and also under RDFS entailment. But Pat says you're free to make extensions. I thought we wanted to indicate which extensions were being used by the presence of URIs, but that wont work in this case, since it's still rdf:first. <TimB-L> but anyway. <TimB-L> Should I send in in a comment on that document? <sandro> It's too late for that, as I understand it. <sandro> As I understand it, unless there is outstanding dissent from before-deadline comments or within the WG, there's nothing anyone can do to stop this from going through to Rec. <sandro> or the WG can re-open the issue in the face of new evidence. <TimB-L> We can annotate the rdf schema with a pointer to others <TimB-L> Like a pointer to a rules! <TimB-L> Like a pointer to rules! <sandro> I can't figure out whether the WG missed important evidence here about how extensions/ontologies were supposed to work, or not.... <TimB-L> { ?x rdf:rest ?y } => { ?y a rdf:List }. <TimB-L> Or whether they got lost in the rest of the document. <TimB-L> The semantics in the document are very weak ly written. "intended to be used ...' <sandro> The question here, I think, is who can say: { ?x rdf:first ?y, ?z } => { ?y owl:sameAs ?z } aka rdf:first a owl:FunctionalProperty. <danbri> anyone. but who would be believed? * danbri looks forward to digitally-signed W3C specs <sandro> danbri, if you said that, would I know you were lying? rdf:first is NOT defined by the W3C specs to be an owl:FunctionalProperty. <danbri> yet <sandro> in rdf 3.0? <danbri> could be. <sandro> that would be an incompatible change, of course. <danbri> that would be consistent with the way things have evolved <danbri> we have model of 'same thing, more claims about it' <sandro> But the more claims about it may render someone else a liar. That's not fair. <sandro> esp when we told them they were free to make more claims about about it themselves. <danbri> I'm happy to deploy on top of RDF. A liar or a partial truth-teller. Since the earlier claims weren't expressed in a machine-way, nor was their implicit scoping, so it's all out of machine's reach anyway. <TimB-L> I think the question would be, who owns the URI? <TimB-L> And w3.org owns it and so we can settle it. <TimB-L> We have a process. <TimB-L> One working group can in fact continuet the work of another. * DanC encourages technical discussion to move to #rdfig or some other googleable space <sandro> And our LC draft settles it. It says user "may" use rdf:first in all sort of other ways. * TimB-L wishes we could stream the tape of this one onto #rdfif without a big mess <DanC> mailing it to www-archive makes it googleabel * sandro would be happy with having this logged and permitting sections of the log. <sandro> yeah * DanC prepares for rdf calendar chat... <danbri> anything i've said today here, am happy -> google * TimB-L would be happy with this being public, esp cleaned up. <danbri> hmm anybody/somebody/nobody'll do it... <TimB-L> W3C owns the IRI. The OWl working group is refining/extending the work of the RDF Core group. <TimB-L> s/IRI/URI <sandro> I'd rather run the problem into the ground before google even gets around to it, myself. <sandro> but I understand the concern that that wont happen. <sandro> So the W3C is free to change rdf:first into being a FunctionalProperty at some point in the future, even though that may break applications written by someone adhering to our specs? <TimB-L> rdf:first I belive is a functional property. That is what the group intended and what people expect. Just that RDF core was not in a position to say so because they didn't know how to specify conformnace to a small set of expected entailments wseparately from the intended full meaning of the term.
Received on Wednesday, 25 June 2003 11:55:06 UTC