- From: Pat Hayes <phayes@ihmc.us>
- Date: Wed, 22 May 2013 17:21:24 -0500
- To: "Peter F. Patel-Schneider" <pfpschneider@gmail.com>
- Cc: RDF WG <public-rdf-wg@w3.org>, Ian Horrocks <ian.horrocks@cs.ox.ac.uk>
Yes indeed, most ingenious. We could phrase this as the number of names in the skolemization of the entailing graph, I guess, or the number of distinct referring expressions in the syntax, but its probably not worth trying to state it the second way. I will try re-drafting that appendix. BTW, do you have any insight onto whether _:x rdf:type ddd . is RDFS valid when ddd is a recognized datatype? In effect, this outlaws datatypes with empty value sets. This seems to me to be a reasonable assumption, but I wonder if it is universal. Pat On May 22, 2013, at 9:43 AM, Peter F. Patel-Schneider wrote: > > There is the claim in the current version of RDF 1.1 Semantics (and > elsewhere, I think - pointers into the literature would be helpful) that > for several variations of RDF entailment, it > suffices to consider interpretations where the size of the domain is at most > the number of names (IRIs plus literals) plus one. This is not correct. > > Consider, for example, the following situation: > 1/ Pick n > 1, m > 0 such that nn = m. > 2/ Pick n distinct IRIs, I1, ..., In, without any special significance in > any RDF-related entailment. > 3/ Pick m-n distinct bnodes, bn+1, ..., bm. > 4/ Let Ni be Ii for 0<i<=n; bi for n<i<=m. > 5/ Construct the RDF graph G containing the following triples > <Ni,Ij,Ik> > for each 0<i<=m and each 0<j,k<=n except when j+n(k-1)=i, i.e., each Ni > has m-1 out of m triples for which it is the subject, but each Ni is > missing a different combination of the predicate and object. > 6/ Construct G' as the following triples > <b,Ij,Ik> > for b distinct from each bnode in G and for each 0<j,k<=n > > Then G does not entail G', as can easily be seen in a Herbrand-like (i.e., > also adding domain elements for each blank node) interpretation for G. > However, in an interpretation with fewer than m elements some No and Np with > o/=p must end up denoting the same domain element, which then acts as the > subject of facts for every combination of Ii as predicate and Ij as object > (speaking a bit loosely here). In such interpretations G' is thus > true, demonstrating that in some cases interpretations with domains of size at least > the number of IRIs plus the number of blank nodes in the entailing graph > must be considered. > > Literals from uninterpreted datatypes can also increase this lower bound. > > When recognized datatypes are present the above constructions can be > relativized to each datatype, providing a bound on the minimum interpreted > size of large (particularly infinite) datatypes that must be considered. > > > peter > ------------------------------------------------------------ IHMC (850)434 8903 or (650)494 3973 40 South Alcaniz St. (850)202 4416 office Pensacola (850)202 4440 fax FL 32502 (850)291 0667 mobile phayesAT-SIGNihmc.us http://www.ihmc.us/users/phayes
Received on Wednesday, 22 May 2013 22:21:53 UTC