- From: pat hayes <phayes@ai.uwf.edu>
- Date: Fri, 19 Jul 2002 16:43:27 -0500
- To: Drew McDermott <drew.mcdermott@yale.edu>
- Cc: www-rdf-logic@w3.org
> [Brian McBride] > With reference to Drew's message: > > >http://lists.w3.org/Archives/Public/www-rdf-comments/2002JulSep/0034.html > > Consider datatypes test A1 with answer YES and literals untidy (i.e. A2 and > A3 NO). > > We have: > > <bag1> rdf:_1 "10" . (1) > <bag2> rdf:_1 "10" . (2) > |= > <bag1> rdf:_1 _:l . (3) > <bag2> rdf:_1 _:l . (4) > >I don't see why (1) and (2) entail (3) and (4). Brian has not published the full gory details behind his example. Allow me to fill them in, so you can better appreciate the reasoning involved here. But be warned, before reading further, that this may corrupt your intuitions so badly that your opinions on the test cases may become worthless. OK, if you decide to read on: We (the WG, that is) found ourselves caught unable to come to a clear decision between the two extreme positions of complete tidyness (literals are unique, have a fixed denotation) and complete untidyiness (literals can mean anything, and different occurrences of the same literal can mean different things.) So we have been trying to find intermediate positions which capture the best of both worlds. One of these positions (we have tried many of them) is the following: literals are syntactically tidy (ie like urirefs, there can be only one node per literal per graph) but semantically untidy, in that the way that the literal is interpreted might depend on the triple in which it is considered; in particular, it might be a function of the pair consisting of the literal itself and the property of the triple. This would allow examples like Brian's A example to have the same literal 'meaning' two different things in two triples, and it allows for 'range datatyping' of properties. Brian however noticed that in two cases there seemed to be very much stronger arguments against it, in that it seemed to produce inferences which are at best highly unintuitive and at worst incoherent. Those two cases involved reification and container membership, where the RDF rules themselves mandate that certain properties be used (in the above case, the rdf:_1 and rdf:_2 membership properties) and hence would impose a fixed interpretation of the literal, even when one would wish to assume that the 'real' property which determines the literal meaning would be some other property (eg in the case of containers, for example, it might be the property which has the container itself as object in a triple.) Hence examples (and reasoning) like the ones given above. Brian's conclusions follow within this 'hybrid' approach to literal meanings. > We've agreed that >literals without datatyping information could mean anything at all. That is the 'extreme untidy' position, which indeed does not support so many inferences. >So how can we conclude that the first "10" denotes the same thing as >the second "10"? > > We have a cardinality constraint of max 1 on rdf:_1. > > Now add some new triples: > > <bag1> rdf:_1 _:a1 . > _:a1 <foo:decimal> "10" . > > This is consistent with (1) above and the cardinality constraint, >and also add: > > <bag2> rdf:_1 _:a2 . > _:a2 <foo:binary> "10" . > > This is consistent with (2) above and the cardinality constraint. > > All together the added statements are consistent with (1) and (2) above, > but not with with (3) and (4) above. > >Yes, they're inconsistent, which shows that the entailment was incorrect. Well, sure; but that entailment was indeed valid in the proposed MT for literals (which Brian did not elucidate on the broader mailing list for fear of letting loose a never-ending debate about alternative model theories....oh well.) Pat -- --------------------------------------------------------------------- IHMC (850)434 8903 home 40 South Alcaniz St. (850)202 4416 office Pensacola, FL 32501 (850)202 4440 fax phayes@ai.uwf.edu http://www.coginst.uwf.edu/~phayes
Received on Friday, 19 July 2002 17:42:53 UTC