- From: Patrick Stickler <patrick.stickler@nokia.com>
- Date: Sun, 16 Jun 2002 15:39:23 +0300
- To: Pat Hayes <phayes@ai.uwf.edu>, Jonathan Borden <jonathan@openhealth.org>
- CC: RDF Core <w3c-rdfcore-wg@w3.org>
On 2002-06-14 3:08, "ext patrick hayes" <phayes@ai.uwf.edu> wrote: > > Here's a rough draft of what Id like to say in the RDF MT document > about 'reserved' (we don't say 'dark' these says) vocabulary, to give > you an idea of what is being proposed here. > > ------ > What does it mean to assert an RDF graph? The normal answer is that > each triple can be read as a simple proposition, and the graph as a > whole represents the conjunction of all of these propositions, so > that what is asserted is the content of all the triples in the graph. > Asserting a triple amounts to saying that it is true, and what that > means, in turn, depends on what defines the meanings of the terms > used in the graph. Before discussing that in more detail, we first > note that it is also possible to use RDF triples simply as a > data-structuring mechanism for encoding expressions of other > languages which have a more complex syntax. If those 'encoding' > triples are regarded as assertions in the same way as other triples, > complexities can arise because the meaning they would have when seen > simply as RDF assertions might not correspond to their intended > interpretation in the other language. To accommodate such encodings > and avoid these complications, we allow that some urirefs may be > declared to be 'reserved'. Triples using urirefs from any reserved > vocabulary can be present in an RDF graph but do not themselves make > any RDF assertions. They may, however, be part of an encoding of > expressions in some other language which itself may be asserted by > the RDF graph in question, according to the semantic rules of that > other language. We note that an RDF parser or processor is not > required to treat such triples in any special way, unless it also > needs to access the content expressed in that other language encoded > in an RDF graph. How does an RDF only processor know which URIrefs are reserved, so that it knows which triples are not asserted? Would we not need a fully generic mechanism to signal non-assertion? > Since reserving a vocabulary effects the meaning of RDF, the > authority to declare a uriref or urirefs 'reserved' in this sense > rests with the W3C. A uriref or set of urirefs is reserved only if > it is declared to be so by a W3C Recommendation. That seems a bit partisan to me. Are only W3C languages layered on top of RDF allowed to be used with RDF? I hope not. > In particular, > reserving a vocabulary cannot be done by simply asserting on a > webpage that it is to be considered reserved. There is no way to > state in RDF, or any language encoded in RDF, that a uriref is > reserved, or for any RDF document to entail this as a consequence. Bugger. Then every single SW application that only cares about RDF assertions has to know *natively* which upper layer language terms are reserved in order for it to determine the truth asserted by an RDF graph?! Surely I must have missed something here, since that is ... er, ahem (I'll be nice and not say out aloud) > ----- > > And then the basic MT rule for I(E) is slightly modified so that it reads: > > If E is a triple S P O . then I(E)=true if S, P and O are not reserved and > .... > > ----- > > That is all that is being suggested. And yes, this is the old > 'unasserted triples' idea in a slightly updated form. If it is not clear to all RDF applications using generic, vocabulary netural means which triples are or are not asserted, then I oppose this proposal on the grounds of it being completely impractical and unscalable for real-world applications. If I write an RDF application that encounters OWL "reserved" vocabulary, I *refuse* to be required to intern special knowledge of that reserved vocabulary in my application, which *never* uses OWL terms, simply to know to ignore those triples and not consider them asserted. The distinction between asserted and unasserted triples must be syntactic and explicit for *every* triple. It is not even sufficient to define some class such as rdfs:Reserved and declare each reserved URIref to be a member of that class, since there is no guaruntee that a SW agent that recieves a triple with such a "reserved" URIref will also possess the relevant class membership knowledge for rdfs:Reserved. So, if Lbase layering and unasserted triples cannot be captured with a syntactic flag without introducing nonmonotonicity, then it would appear that layering itself cannot be done monotonically. Regards, Patrick -- Patrick Stickler Phone: +358 50 483 9453 Senior Research Scientist Fax: +358 7180 35409 Nokia Research Center Email: patrick.stickler@nokia.com
Received on Sunday, 16 June 2002 08:35:08 UTC