- From: Axel Polleres <axel.polleres@deri.org>
- Date: Wed, 04 Jul 2007 13:43:21 +0100
- To: Jos de Bruijn <debruijn@inf.unibz.it>
- Cc: Dave Reynolds <der@hplb.hpl.hp.com>, RIF <public-rif-wg@w3.org>
Jos de Bruijn wrote: >>>>>What we cover is well defined. We cover simple entailment, RDF >>>>>entailment, and RDFS entailment. See [1]. >>>> >>>>As I said in [*] >>>> >>>>(a) I don't think we need claim that the rules perform simple entailment >>>>at all; >>>> >>>>(b) I would rather that we defined one or more rulesets and make those >>>>rule sets themselves the specification of what is covered. I.e. they are >>>>not "implicit" but "explicit". In particular, different RDFS >>>>applications typically make use of different subsets all of which are >>>>easily handled by rules. We would bless a couple of standard rule sets >>>>(giving them standardized URIs so that importing of those rulesets can >>>>be hardwired into RDF-aware RIF processors). >>> >>>I must say that I do not really like this approach, because it is hard >>>to know what is going on exactly if we just give some rule sets which >>>define some kind of semantics for RDF. >>>Furthermore, the RDF specification defines three normative kinds of >>>entailment. I would say that they are normative for a reason; we should >>>not ignore the specification. >> >>we shouldn't, but it might make sense to define some rulesets which e.g. >>do not include all the infinite axiomatic triples. > > > We can certainly demonstrate some cases in which these axiomatic triples > are not necessary for reasoning. > > >>Apart from that, what you write in [1] pretty much makes sense to me, >>We could summarize what we seem to agree so far as follows: >> >> - treat bnodes in a head by skolemization (or the built-ins suggested). > > > I assume you refer to the heads of N 3-like rules? > It is, at the moment, not entirely clear to me how they should be treated. > > >> - treat bnodes in a body or query as (special) variables. > > > Right. > > >> what yet seems to be open for me is three things: >> >> - how we'd deal with bnodes in N3 like rules reused in several rules. >> Alternatives: - forbid >> - standardize apart upfront. >> - other ... ? > > > I am not sure what you mean with "apart upfront" . I mean "standardize apart" = "disambiguate by renamiing" > I have no idea why > we would want to forbid their reuse. What is the semantics/scope if they are realy reused? An existential variable in a rule body is scoped existentially over the rule, not over the ruleset. The semantics of "reuse" would be very unclear to me, and I don't know whether we'd want that. Imagine: _:a p o. s q o. :- _:a q o. _:a r o. :- ?x q o. Is the _:a the "same" blank node in all three cases, or treated differently? > the question is really whether bnodes with the same name in different > rules refer to the same thing, or not. I rather hope the latter. yes, so we'd agree the above would be equivalent to: _:a p o. s q o. :- ?a q o. sk(?x) r o. :- ?x q o. ??? >> - how we'd deal with bnodes occurring in both head and body of N3 like >> rules.Alternatives: - forbid >> - treat both as variable >> - treat hewad as skolem and body as variable >> - other ... ? > > > Again, I hope they are interpreted differently. That is fair enough in my opinion. >> - how we'd deal with results of rule(set)s? ie. if we have a frame >> s[p->o] in the head of a rule (where s,p,o is a variable skolem >> term, literal, iri, etc), it might well not be a valid RDF >> triple, so my question is, for N3 like rules, how would we ensure >> that the result is getting back to RDF again. Do we need/want that? >> >> (Here, e.g. SPARQL takes the approach >> to simply ignore all non RDF triples, and only consider valid RDF >> triples as consequences) > > > Here, again, SPARQL uses an ugly syntactic means for getting rid of > non-well-formed RDF: triples with literals or bnodes in inconvenient > places are simply discarded. I would argue that we do not want any such > arbitrary filtering in RIF. Hmmm, but what's the alternative? > In any case, I do not think it is up to this working group to decide how > people should construct RDF graphs from query answers obtained using > rules exchanged with RIF. If we see the interoperation with RDF only one-way then yes, but I am not sure whether all of the WG like that (I myself would, apart from possible technical difficulties love to have the possibility to export rules results/models/inferred atoms back to RDF) best, axel -- Dr. Axel Polleres email: axel@polleres.net url: http://www.polleres.net/
Received on Wednesday, 4 July 2007 12:43:38 UTC