- From: Geoff Chappell <geoff@sover.net>
- Date: Sat, 2 Jun 2001 07:47:10 -0400
- To: "Peter Crowther" <peter.crowther@networkinference.com>, "'Graham Klyne'" <GK@ninebynine.org>
- Cc: <www-rdf-logic@w3.org>
----- Original Message ----- From: "Peter Crowther" <peter.crowther@networkinference.com> To: "'Graham Klyne'" <GK@ninebynine.org> Cc: <www-rdf-logic@w3.org> Sent: Saturday, June 02, 2001 8:09 AM Subject: RE: rdf as a base for other languages > > From: Graham Klyne [mailto:GK@ninebynine.org] > > At 01:07 PM 6/1/01 -0400, Peter F. Patel-Schneider wrote: > > [...] > > >Sure, you can do anything you want outside of RDF. However, > > >if you want > > >RDF to represent anything, you better do all your work within RDF. > > > > Why the insistence on all-or-nothing? Is there any > > fundamental reason why > > we cannot start with a language capable of expressing ground > > facts, and > > extend it in a consistent way (creating a new language, "outside" the > > original) such that the original language for expressing > > ground facts is present as a sub-language? > >[...] > > With RDF as it stands: > > - If you posit that the original language allows such extensions because > it's all just triples, then there's no way to denote what is original versus > what is extended and there's therefore no way for an agent that only > interprets the original language to know when its fallen off the edge of a > statement in the original into an extension --- that's fine if everyone uses > the same extension language, but then why is RDF around at all because the > separate layer of RDF is not useful on its own. So everyone can use (say) > DAML+OIL, but you cannot rely on any information a non-DAML+OIL agent > extracts from that chunk of RDF that uses DAML+OIL because it may make > incorrect assumptions about the extensions: for example, that the relation > petersExtension:negation does (or does not) represent the concept of > negation. > > - If you posit that the original language does not allow such extensions, > then there is no way of representing anything other than positive ground > triples and the language is of very limited use. So you can't create > DAML+OIL on top of RDF, and have to encode it in some other way. > > For me, the minimum addition to RDF would be a way of distinguishing > 'traditional' RDF positive ground triples from triples that have been > generated as structure by some language outside RDF. Ideally I'd like to be > able to distinguish *which* language generated the structure as well. > > I wonder whether the RDF aficionados are assuming that this is done using > namespaces, and that rdfs:* is the 'original language' and everything else > is an extension? But if so, the language is even more limited as you cannot > create your own subjects, predicates, and objects; it seems that something > outside the namespace mechanism is required. > > - Peter It seems to me RDF as a method of exchanging positive ground triples between agents of varying capabilities will be useful regardless. There will surely be consumers of RDF that have no inference capabilities whatsoever and will need to rely upon producers of RDF to make whatever inferences they are able to make (or choose to make) before returning results as ground facts. This may result in a lossy knowledge transfer as the generalized intentions encoded in the producers knowledge base are "hardened" into specific facts acceptable to the consumer or are disregarded altogether, but you can imagine that this will suffice for many applications. And as you say, any agent should be able to distinguish facts encoded for a particular extension and disregard them (or at least be able to avoid misinterpreting them). It only seems reasonable that any number of different reasoning systems will be build to work with RDF with many accepting the various logical restrictions of prolog, datalog, etc. in exchange for performance. Surely each of these systems can encode its own language as rdf triples. Even if it means that only a similar system can interpret that language, any system can interpret the output of the system (as long as it is conforming ground triple rdf). And if each of those systems included a description of its reasoning system in kif, we would have the basis for interchange of other than ground facts between systems. Is that unreasonable? unworkable? Geoff
Received on Saturday, 2 June 2001 09:48:46 UTC