Re: New RDF model theory (well, damn nearly)

>Pat--
>
>You wrote:
>
> >PS. Frank, I still havnt considered the issue you raised about 
>rdf >vocabulary having to have an 'associated' schema, because I
> >now realise I have no idea what it means, or at any rate how it 
>would >effect the semantics. (Suppose there wasn't an >'associated' 
>schema? >What RDF graphs would change in meaning? )
>
>I think this "associated schema" business is not as precisely 
>expressed as it might be in the M&S, but the following musings try 
>to convey the general idea I had in mind (and we can possibly take 
>this discussion as an example of using the model theory to clarify 
>what the specs ought to say or not say).  I don't think any of this 
>will be unfamiliar to you, since DAML is based on the same idea, and 
>I don't think the DAML model theory covers this issue explicitly 
>either (so maybe it's inappropriate in a model theory).  Anyway:
>
>Background:
>
>As I said in my original message, there are various places in the 
>M&S that tie property names to specific schemas:
>
>*  p60 says "Property names must be associated with a schema."  It 
>then goes on to say that this can be done by qualifying the names 
>with a namespace prefix to identify "the corresponding RDF schema" 
>[admittedly a *syntactic* mechanism].
>
>*  Section 2.2.3 further discusses the idea that the meaning of a 
>property name will be pinned down by associating it with a 
>particular schema.  In particular, p85 says "Meaning in RDF is 
>expressed through reference to a schema."  [Hence the relationship 
>with semantics].  This paragraph also muddies the waters a bit by 
>saying "A variety of schema forms can be used with RDF, including a 
>specific form defined in a separate document [RDFSchema]..."
>
>The general idea, then, is that people will identify the intended 
>interpretation of the property names they use by defining a schema 
>containing them, along with associated definitions and constraints 
>that reflect (to the extent the schema capabilities allow) that 
>intended interpretation (that intent can be documented "more 
>meaningfully", as far as computer processing is concerned, by using 
>more powerful languages, e.g., DAML+OIL vs. RDFS, but that's 
>something of a separate issue).  The reference to a schema 
>distinguishes a particular property name from the same name used 
>with a different meaning by somebody else, possibly defined in a 
>different schema.
>
>My original thought, specifically due to sentences like "Meaning in 
>RDF is expressed through reference to a schema", was that this 
>seemed like something that ought to be reflected explicitly in a 
>model theory of RDF.
>
>Development:
>
>Going through the model theory (especially before you added coverage 
>of RDFS), I think I see what you mean about this not necessarily 
>having any effect on the semantics.  [Interpolation:  your question 
>"Suppose there wasn't an 'associated' schema? could on one hand be 
>answered by saying that it's syntactically irrelevant:  the M&S says 
>there must be an associated schema.  However, the question needs to 
>stay on the table, because it may be something that the M&S doesn't 
>need to say, or needs to say in some other way.]  That is, as far as 
>the RDF graph is concerned, properties are URIs, are thus 
>unambiguous, and have the meaning of the associated resource (in the 
>case of properties, presumably the meaning attached to the resource 
>by whoever created it).  This is reflected in the MT by the URI to 
>resource mapping (for resources that are properties).  The semantics 
>would be essentially unaffected even if whoever created and used a 
>property didn't explicitly document the meaning they had in mind for 
>it anyplace (problems might arise if there was a need to share that 
>meaning with anyone else, but we're not talking about that yet.) 
>Taking this line, the reference to an associated schema is just a 
>way of saying that there is a need for some *syntactic* mechanism to 
>allow RDF writers to universally identify a specific property, and 
>referencing a schema provides a way to do that.
>
>On the other hand, there remains this idea, mentioned above, that 
>people will try to describe the intended interpretation of property 
>names by defining schemas.  The schemas may contain additional 
>information about the property resources identified by property URIs 
>(like constraints, which constrain the valid instances of the 
>property).  This seems clearly related to the semantics of the 
>associated graphs.  The intended usage is that people will define 
>schemas that reflect their intended interpretations for the RDF they 
>write, and then write the RDF.  The RDF will be indirectly 
>asssociated with its semantics via the reference to the schema. 
>Now, this isn't universal, in the sense that it covers all the RDF 
>that may be written.  For example, even if there were an RDF schema 
>that you and I explicitly refer to in exchanging information, we 
>could also have some "side agreements" where we use additional 
>terms, not reflected in the schema, whose meaning only you and I 
>knew and agreed to.  However, a general idea in RDF (and more 
>extended Semantic Web) usage is that meanings will be grounded by 
>associating terms with schemas (or, by extension, ontologies).  What 
>I was after was something in the model theory that explicitly 
>reflected this idea.  I haven't yet digested your MT section on RDFS 
>interpretations enough to tell if there is anything along these 
>lines in there (is there?), so this response may be premature (or 
>just laziness on my part;  asking you to tell me, rather than 
>figuring it out for myself!).  It seems to me that there might be 
>something, for example, that explicitly tied the interpretations of 
>the classes and properties defined in the schema to the 
>interpretation of, e.g., those properties when used in the graphs. 
>It seems to me that the situation we have here is somewhat different 
>from the situations in which model theories are used in "vanilla" 
>FOL, in that here we have some explicit definitions in schemas (and 
>ontologies) that attempt to, if only very partially, specify or 
>constrain the intended meaning of the names that are used, and that 
>somehow the model theory should somehow reflect that "role" that the 
>schemas are trying to play in defining the meaning of associated 
>graphs.  Does this make any sense?

Yes, it does, (thanks) but Im not sure how to make it concrete. The 
MT, like all MTs, just restricts itself to specifying what 
expressions denote and when they are true (in an interpretation); it 
doesn't really have any way to distinguish which parts of a formalism 
play a certain 'role', beyond their syntactic role in other 
expressions. As far as I can see, the only way that an RDFS schema 
specification can impose itself on a part of the RDF vocabulary is by 
specifying some classes (ultimately, as Dan C. points out, by using 
rdf:type) and then using rdf:range and rdf:domain, and these have a 
semantics already, so what more can we do?

Pat

---------------------------------------------------------------------
(650)859 6569 w
(650)494 3973 h (until September)
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes

Received on Monday, 27 August 2001 15:47:13 UTC