- From: Dan Brickley <danbri@w3.org>
- Date: Thu, 25 Apr 2002 14:27:12 -0400 (EDT)
- To: Jeff Heflin <heflin@cse.lehigh.edu>
- cc: <www-rdf-logic@w3.org>
symbol grounding problem: quite! When people ask me whether RDF has been taken over by the AI crowd, I of course say "no, of course not!". The more up front we are about admitting all this stuff is ultimately grounded out by prose definitions, by humans for humans (including lawyers, logicians etc), the happier I am. You're quite right to note that there won't be a single set of 'primitive classes' that underpin the rest of the logically precise SemWeb. The scruffy bits of prose we shove in rdfs:comment and rdfs:label properties are pretty important when it comes to fixing the real world meaning of RDF/XML instance data. As are supporting documents, examples etc. The phrase 'garbage in, garbage out' applies pretty well to the Semantic Web: if we don't get the prose definitions right, then all the model theories and axioms in the world won't make our ontologies and rdf documents meaningful... Dan On Thu, 25 Apr 2002, Jeff Heflin wrote: > My personal opinion is that if you're using an ontology language, every > term you use must be defined in some ontology (even if only to say that > it is a class or property). Now many, including members of the Joint > Committee and the WebOnt WG, disagree with me. They often argue by > analogy to the Web dropping the need for all hypertext links to actually > point to something. This allowed the Web to scale beyond previous > hypertext systems by simply introducing 404 errors. However, I don't > think the analogy quite works here. The whole point of an ontology is to > achieve some standard agreement on the meaning of terms. If you can just > create a term out of the blue, then what's the point of having an > ontology in the first place? > > Still, as you point out, somewhere the system will have to bottom out in > terms that do not have formal definitions. In description logics, they > call these primitive classes. The Semantic Web will have to have > primitive (undefined) classes (or else solve the symbol grounding > problem (ha!)), but I think that these should come from a set of terms > defined by some community (in an ontology). Now, it is not necessary > that everyone build their definitions from the same set of of primitive > classes (that wouldn't scale!), but it is important that when two > different domains need to be integrated, that it is possible to create > ontologies containing axioms that bridge (at least partially) between > the classes defined by the ontologies for these domains. > > Jeff
Received on Thursday, 25 April 2002 14:27:13 UTC