- From: Graham Klyne <GK@ninebynine.org>
- Date: Fri, 31 May 2002 11:24:09 +0100
- To: patrick hayes <phayes@ai.uwf.edu>
- Cc: <www-rdf-comments@w3.org>
At 02:42 PM 5/30/02 -0500, patrick hayes wrote: >>> Yes, but my point is that this logical interpretation is *not* RDF >>> entailment. It is, instead, RDFS entailment. >> >>I still fail to see why it is important for us to classify entailments. >>It's just going to over complicate stuff needlessly. If my agent knows the >>rules for rdfs:subClass, than it can arrive at legitimate entailments, > >There is no universal overriding notion of 'legitimate' entailment, is >why. I agree it complicates things, but I see no way around the fact that >life is complicated. I agree - there can be no universal entailment, except in the trivial sense outlined below. Further, this discussion makes me think that classification of entailment types is precisely something that should be considered. I can imagine a class (not rdf:Class) of rdf:Properties whose intended interpretation is to relate N3-formulae resources according to different entailments; e.g., using N3: { ant-graph } ent:no-entailment { con-graph } . -- is always false { ex:x rdf:type ex:foo } ent:rdf-entails { [rdf:type ex:foo] } . -- is true { <ex:Jane> <rdf:type> <ex:Woman> . <ex:Woman> <rdfs:subClassOf> <ex:Human> } ent:rdfs-entails { <ex:Jane> <rdf:type> <ex:Human>. } -- is true, but { <ex:Jane> <rdf:type> <ex:Woman> . <ex:Woman> <rdfs:subClassOf> <ex:Human> } ent:rdf-entails { <ex:Jane> <rdf:type> <ex:Human>. } -- is false { ant-graph } ent:foo-axiom-entailment { con-graph } . -- is true if some specification or authority 'foo' axiomatically asserts the entailment. { ant-graph } ent:universal-entailment { con-graph } . -- is always true Under the rdfs:subProperty relation, treated as a partial ordering, these entailments would form a lattice with ent:universal-entailment as 'top' and no-entailment as 'bottom'. Further, I think one would wish to require that ent:rdf-entailment is a subproperty of all other entailment properties. Maybe also ent:rdfs-entailment (other than of ent:rdf-entailment), so we end up with a lattice something like this: ent:universal-entailment | . . . . [ various other flavours of entailment ] . . . . ent:rdfs-entailment | ent:rdf-entailment | ent:no-entailment Of course these are all just intended meanings -- there's no way to logically deduce them using just RDF/RDFS or even full FOL. But what this does is suggest a vocabulary that allows agents to indicate *how* they arrive at certain conclusions, and acceptance of such conclusions can be tempered by whether other agents agree that the method used is valid for their purpose. I think this makes the limited scope of RDF-entailment (and RDFS-entailment) quite explicit, provides a hook for linking RDF inference to the anticipated web-of-trust developments, and also provides an explicit marker for noting inferences that are the result of some logical reasoning as opposed to those which result from some non-logical (and sometimes illogical) reasoning including things like "because it's the law". Thus: { :I ex:owe [ ex:to :U ; ex:amount "1" ; ex:denomination UScur:dollar ] } ent:US-money-entails { :I ex:owe [ ex:to :U ; ex:amount "4" ; ex:denomination UScur:quarter ] } #g ------------------- Graham Klyne <GK@NineByNine.org>
Received on Friday, 31 May 2002 07:15:56 UTC