RDF issues: schema and inference

I've just taken a long-overdue peek at SHOE, and (if I understand right) it 
closely parallels some thoughts that have been troubling me about RDF schema.


First, my take on SHOE -- it defines two parts:
(a) a type system that is structurally equivalent to the RDF schema type 
system:  classes with multiple inheritance.
(b) a basic framework for inferring new statements from existing statements 
(including type assertions).


Next, my troubles with RDF schema:

Basic RDF, and RDF schema, generally avoid associating a logic with RDF 
statements.  This is fine and good.  But there are some respects in which 
RDF schema introduces elements that might be regarded as a kind of logic, 
such as:  from (A subClassOf B) and (B subClassOf C) infer (A subClassOf C).

Some similar inferences can be drawn around property and subPropertyOf.

(Also, from some discussions than have been held, there is a school of 
thought (which I support) about range and domain semantics that would allow 
inference of class information about graph nodes from properties that are 
applied to them.)


So why does this trouble me?

On the surface, this inference looks like logical implication, which, as 
TimBL shows in his semantic toolbox paper [1], is pretty much equivalent to 
full first order logic (FOL).  But (and I take this as axiomatic) the full 
force of FOL should not be imposed on the basic RDF and RDF schema layers.

I had been thinking that the way out of this is to make a clear distinction 
between 'inference' and 'implication'.  And SHOE is my existence 'proof' 
that this is reasonable.

Within the RDF core, I think there is no concept of truth or 
falsehood.  Statements just exist, or don't exist.  Inference, then, can be 
viewed as a mechanism for creating new statements from those already to 
hand;  nothing more.  Implication is a logical relation:  A=>B == -A|B, 
which depends upon the truth and/or falsity of A and B, and which is used 
in practice to deduce the truth of B from the truth of A.


And finally, my suggestion:

If/when there is a review of the RDF schema document, I suggest we should 
be clear that some basic inference mechanism(s) are recognized (but without 
truth, falsehood or negation), and the spec should state explicitly those 
inferences (i.e. new arcs) that can be drawn from an RDF graph and/or 
schema graph.

It might also be possible to extend (or build upon) the schema to describe 
inferences, a la SHOE.  I'm not sure whether or not this is a good idea.

#g

[1] http://www.w3.org/DesignIssues/Toolbox.html
------------
Graham Klyne
(GK@ACM.ORG)

Received on Thursday, 30 November 2000 12:32:56 UTC