- From: Wolfram Conen <conen@gmx.de>
- Date: Tue, 20 Nov 2001 12:31:40 +0100
- To: "Peter F. Patel-Schneider" <pfps@research.bell-labs.com>
- CC: www-rdf-interest@w3.org
"Peter F. Patel-Schneider" wrote: > > > My view of this paper is that it introduces a translation of the new RDF(S) > > > model theory into Horn rules with stratified negation and then complains > > > about characteristics of the result. But are these complaints about the > > > new model theory or about the translation? > > > > > > > Neither (by the way: note that the translation of the RDF(S) model > > theory into the eighteen datalog rules does not require stratification, > > though). > > Yes, stratification is not needed, it is just that that is how you describe > the formalism used. > Some form of negation, however, is used. In our old interpretation, stratifiable negation is indeed used (for the integrity constraint checking). In our new interpretation it is only used with the "lit()" (Literal) stuff, which is our suggestion to deal with the "uuu" used in the closures rules (see below for details). I missed out the "not lit(x)" when writing the first answer, so, sorry, yes, stratification is relevant for both interpretions. >... > > It suggests that it might be reasonable to complain about the > > loss of expressivity due to the decision of the RDF Core WG with respect > > to domain/range constraints (ie, allowing to deduce types). > > This is a valid criticism. Different parties can take different stands on > this issue. > > The criticism of treatment of literals, on the other hand, appear to be > much more of a criticism of the translation, but are couched as a criticism > of the model theory. The criticism was simply motivated by the distinction between aaa/xxx and uuu that is used for computing the closure rules in the MT and not justified/explained in complete transparency there. If you allow labels that are syntactically valid (rdf_2396 conform) URI to be used on nodes representing resources and on nodes representing literals, it becomes somewhat difficult to make the aaa/xxx and uuu distinction on the pre-interpretation level used for computing the closure rules (even if you would already resort to XL and IS, they could both map the same "syntactical URI" to different things and you would still be left with the problem to decide about the actual role of the label (xxx/aaa or uuu) if it is used in object position only, for example). Also, saying that an instance of rdfs:Class gives you also a subclass of rdfs:Resource and having the fact that rdfs:Lierals is an instance of rdfs:Class seems to be a bit problematic (though, as has been pointed out in the paper, with avoiding (through the use of the "uuu" stuff) to have explicit representations of Literals being typed as such in the closure graph, the class extension of rdfs:Literals will be empty, and thus the interpretation condition about subset inclusion will be satisfied, but still looks a bit strange). We usually try to be constructive in our criticism (by suggesting some directions that might be taken for solutions, the details are scattered throughout the paper), so I hope the reader of the paper will not be left with the feeling that we do not like the model theory (which is not the case). Best, Wolfram (Peter, if you (or others) should have comments on the details of the paper, you may want to send them directly to us. We will then prepare a summary of discussions/changes to accompany the paper. I just had a small accident yesterday and particpating in a timely manner in public discussions will be a bit difficult for the next time (and Reinhold is currently not available either). Thanks.)
Received on Tuesday, 20 November 2001 05:30:16 UTC