Meaning of URI Refs
(This is a very long and complex message. Please refer to the original message at: http://lists.w3.org/Archives/Public/www-rdf-comments/2002OctDec/0043.html)
I'll try to precis:1. An RDF document must be understood to assert the truth of all documents named in the URI parts of the URIrefs that it uses to label nodes and properties. 2a. The difference between formal and natural language is overplayed. 2b. Predicate has no special status in determining meaning. 2c. Only two kinds of entailment. The rest is, as far as I can tell, a discussion of the issues raised by these claims.
1. RDF document accepts truth of documents indicated by URIrefs it usesI think there's can be big difference in principle between accepting the truth of a document and accepting a definition given in a document. In some cases, if a definition depends on the entire content of a document that contains it, the two may become the same. I'm not ready to go beyond the first level of acceptance. I don't think RDF needs to say any more than that at this time, whatever community consensus about the latter may be. 2a. The difference between formal and natural language is overplayed. Hmmm... maybe. The meaning is still socially present, but cannot be formally inferred. All-in-all, we may be better to pull this last part of section 2.3.3, since I don't think it really adds any new information to what has gobe before. In it's original form, I think it was intended as a clarification, but it seems to be more prone to cause confusion, or disagreement. 2b. Predicate has no special status in determining meaning. This was my interpretation of a comment made by TimBL. I think the statement as made is true: for the most part it reinforces the idea that the meaning of a URI isn't subject to whimsical change. But the emphasis on the predicate may be overstated. I could see if I can soften the language there 2c. Only two kinds of entailment. I think this is more a matter of presentation than of fact. I think I can see why you find RDF-entailment + "the rest" to be appealing, but I think that could be a very difficult message to convey to a wider audience, for whom the very idea of entailment may be a new concept. And I think "no way to formally express allowable inferences" is exactly true of RDF (core). Beyond that, I agree with what you say about informal description of formal terms, but think that would be too much to try and say in this document. --- This is a long commentary, which touches on many subtleties. Most are beyond the scope of this document to define completely, because the issues are not fully understood to the extent that there is a clear community consensus. I find I agree with much of what is said in sections 4 and 5, to the extent that I understand it, and don't see that it is greatly at odds with the current text. Note that I've also added a section 2.4.6 discussing entailment, which I think touches on some of the points raised. Finally, I think a point where we diverge is the idea that RDF entailment is a relationship between documents. I don't think it is. I think it is a relationship between RDF graphs, which may be contained in documents or constructed from the contents of one or more documents. There is some machinery at work here that I think is outside the scope of the core RDF language specification.
Follow-up from Sandro, and ensuing discussion on RDF-interest.The 4th citation (0119.html) contains an opposing viewpoint from Pat Hayes.
I think this debate is reading too much into the RDF core specification.Concerning: Re: RDF entailment is a relationship between graphs, not documents. I don't think the difference matters. I'm happy to rephrase everything I said to be about graphs. (or I would be if it weren't so long!) I do think there's a difference, because a graph has a clear formal definition, but a document does not (in this context). What this means is that if one provides a formal definition of how to construct an RDF graph from a document or documents, then the RDFcore definition of entailment can be applied. Future work may provide such a definition, but to try and do so now would risk the progress we have made. Concerning: Re: "I think [there] can be big difference in principle between accepting the truth of a document and accepting a definition given in a document." Yes, in principle, but probably not in fact (alas). See , seconded by Peter. Well, I think I mis-stated that slightly. RDF itself doesn't do definitions in this sense: it is interpretations that define denotations of a URIref. So one won't find any definitions simply by looking at an RDF graph, just assertions that are true under satisfying interpretations. One might introduce some convention for a publisher to use RDF to define what they mean by a URI, but that would be an additional, extra-RDF convention, and the convention used must distinguish between definitional and other material. Which puts it out of scope for the current RDF specifications.
No specific change has been suggested that we think is appropriate at this time.