- From: Wolfram Conen <conen@gmx.de>
- Date: Tue, 05 Jun 2001 15:24:27 +0200
- To: phayes@ai.uwf.edu
- CC: www-rdf-logic@w3.org
pat hayes wrote: > > > > >Once you start to "interpret" triples with a host formalism, you may > >easily get rid of reification if you interpret it as nesting. > > But how do you tell which reifications are to be interpreted as > nesting and which are to be interpreted as reification? > Because reification, if understood as being, indeed, reification, > is not nesting. > That was the point of my reply to Sandro: what we need are > ways to use subexpressions without mentioning them and without > asserting them. "Nesting" is a good word for this. > I took the following approach: lean back, Wolfram, relax, forget some assumptions about the meaning of some words, read the RDF M&S document and try to come up with a reasonable (workable?) interpretation. And I haven't found too much in the spec that forbids to interpret that what is being called "reification" in RDF M&S as nesting (and only as nesting). I think that the understanding of reification that you mention above, may certainly be useful but requires a proper notion of semantics (while nesting helps to build expressions) - so, for me, it can be tackled as soon as the "expression" layer is settled and we start to talk about the details of semantics. > > > You "only" have to write useful mappings and axioms to deal with > > this nesting. > > In what language? If we had a language to write axioms and rules > which could deal with this stuff, all our problems would be solved. > That is where we came in. > My view was "pragmatic": we wanted to be able to express some semantics "on top of" RDF, to exchange this semantics, to change them if needed etc. so that we could start building some "semantically-enhenced" applications., so we took the following approach: * view RDF M&S as allowing to formulate a set of 3-ary expressions.[With "interpreting" reification as nesting (and, possibly, containers as lists), it could also be transformed into a set of nested 3-ary expressions (and lists)]. * map the 3-ary expressions to a predicate "triple" in Prolog * add a predicate "definedAs" to RDFS that allows to make Prolog-definitions of properties explicit. Express the RDF Schema concepts with rules in Prolog, and augment the Schema documents accordingly. * Add the mapping and the "definedAs" on top of Jan Wielemaker's RDF Parser (SWI-Prolog) and start playing around with "little languages" that are defined in RDF (Schema) documents based on an explicit and accessible axiomatization of properties (for an implementation, see the RDF Schema Explorer [1]). We, then, were able to experiment with semantics of properties/constraints, play around with alternatives, be explicit about our intentions etc. without having to wait for a standardisation to come to a conclusion and the subsequent development of stable tools that implement the standardized semantics. Hm, maybe such an approach -- very few and simple "built-ins", axioms accesible in the RDF Schema documents, "truely" extensible on the grounds of the chosen host formalism (plain Prolog in this case, though XSB with tabbling would sometimes be better suited, for example to axiomatize subProperty) -- is more practical then trying to define yet-another-general-purpose language? I don't know. Well, at least the RDF Schema explorer has already been helpful for Reinhold's further development of XWMF and some access-control stuff, so it seems to be a "workable" approach somehow, and it gives an enjoyable freedom to adapt semantics to the requirements of the domain (in our experience). There is still a huge number of interesting questions: impact of distributed, shared, partially accessible, "secured" "knowledge bases", impact of out-dated, updated, deleted knowledge, etc. which all may become interesting in a Web context, that will probably be more easily discussible if a number of practical experiences with different semantics in different domains have been made (for example, based on the "host formalism" approach from above?). If one would really want to have ONE "RDF-like" language only, why not take OIL/DAM-L which seems to have what some people most urgently want: a clean semantic (and forget about RDF Schema?)? [Not to be misunderstood: if RDF M&S would be cleanly seperated from schema constructs, would precisely and unambigously define, how expressions (RDF models, ideally with nesting/lists) have to be computed from the XML/RDF syntax, and if we would start to standardize some mappings of RDF and axiomatizations of RDF Schema to some host formalisms, I would prefer the "host formalism" approach outlined above, because it gives the freedom to adapt the axiomatization even to exotic requirements whenever the need arises and is, still, in some sense, interoperable (and available)]. Best regards, Wolfram [1] http://wonkituck.wi-inf.uni-essen.de/rdfs (newly updated paper available upon request)
Received on Tuesday, 5 June 2001 09:22:43 UTC