Re: SW, meaning and reference

If I understand this debate so far, a putative "theory of reference" would 
provide a framework through which an arbitrary community (without 
predefined bounds or membership) would have each of its members associate 
the same concept with every name that is used in communications between 

This is tough.  Thus, model theory looks for truths that can be asserted 
independently of a theory of reference.

OK so far?

If so, is it reasonable/useful to think about a partial theory of reference 
which sets out some properties of the relationship between a name and the 
concept(s) it denotes in a Web context, without attempting to define how an 
actual relationship is determined?

I am thinking that such a partial theory might, for example, describe 
properties of such relationships in a Web/Internet environment, maybe 
capturing ideas like:
   What are the observable properties of a "resource"
   What can be said about the relationship between a URI and "resource"
   What, if anything, is the difference between a "resource" as used by
     RDF and a resource as used by the Web in general
     (esp w.r.t. fragment identifiers)
   Certain operations, such as HTTP GET, do not have side effects
   Other operations may have side effects;  characterization of such


At 03:00 PM 8/31/01 -0400, Dan Brickley wrote:

>(happily removing QNames from Subject: line)
>On Fri, 31 Aug 2001, pat hayes wrote:
> > >On Thu, 30 Aug 2001, pat hayes wrote:
> > >
> > > > [...] Really pinning down actual *reference* to *real things* is a
> > > > very, very  tricky business, one that is way beyond the ability of
> > > > current semantic theories to analyse, and therefore probably best
> > > > left aside for now in these discussions.
> > >
> > >Amen! This is however something that Semantic Web (via our use of both
> > >URIs and of RDF descriptions for identification) will need to
> > >engage with at some point.
> >
> > I agree, but be ready for having to do some hard, basic (and
> > therefore slow) research. New syntaxes are easy, and model theories
> > can be churned out to fit most any reasonable intuitions about
> > consistency, but reference is harder.
>Yes... reference is far murkier, theoretically. And imho likely to stay
>that way. The precision that we achieve through DAML+OIL-style formalism
>will always be in tension with the unavoidably fuzzier aspects of meaning
>associated with our notion(s) of reference. (which is fine; interesting 
> > > I do think that the 'reference' aspect to
> > > 'meaning' is something we'll need to deal with if SW is to fulfill
> > > expectations w.r.t. ecommerce etc.
> >
> > Ecommerce only requires something that works pragmatically.
>Sure; I don't mean to suggest that the SW will be paralysed because nobody
>has perfected a formalisation of how URIs denote. People will find way to
>build systems that buy and sell (and lie and cheat) regardless.
> > I'm sure there are several ways to make the SW work well enough for 
> ecommerce,
> > but if the $$ have to wait for a universal theory of reference, the
> > investors had better be ready to take a very long view.
>Wiser to give up on trying; everything I've read on theory of reference
>inclines me to think that the only way to succeed is to give up.
>Stephen Stich writes (imho) quite persuasively on this...
>(not quite the excerpt I was looking for, but all that Google could find...)
>Before I could start on that project, however, there was a prior question
>to be confronted. If the goal was to produce a correct theory of 
>reference, I would have to
>get clear on what it is that makes a theory of reference correct or
>incorrect. What exactly are the facts that a correct theory of reference 
>is supposed
>to capture? And how can we find out whether a theory has succeeded in
>capturing those facts?
>That argument starts with a hunch, albeit a widely shared one. While there
>are lots of theories of reference on the market  these days, my hunch is
>that the accounts that do the best job at capturing
>people's relatively firm and stable intuitions about reference are not  those
>that follow the path staked out by Ramsey and Lewis, but rather those that
>tell what Lycan calls a "causal-historical" story. The basic idea
>of these theories, as we saw earlier, is that words get linked to things
>in the world via causal-historical chains. The first step in creating such
>a chain is a "grounding" or a "reference fixing" - an
>event or process (or, more commonly, an array of such events or processes)
>in which a term is introduced into a language to designate an object or
>a kind of objects. Following this there is a series (often a very long
>of reference preserving transmissions, in which the term is passed from
>one user to another, preserving the reference that was fixed when the term
>was introduced. But, of course, not just any way of introducing a term
>into a language will count as grounding the term on a particular object or 
>of objects, and not just any way of passing a term from one user to
>another will count as a reference preserving transmission. The legitimate
>groundings and transmissions will be those embedded in causal-historical 
>chains that
>are sanctioned by intuition. When one looks carefully at the class of
>groundings and the class of transmissions that pass this test, however, it 
>that in each class the allowable events are a mixed bag having at best a
>loosely knit fabric of family resemblances to tie them together.
>While I could imagine someone setting out to redescribe URIs in terms of
>an initial 'GroundingEvent' or 'NamingEvent', and (say) a
>causal-historical account of naming, I wouldn't join a W3C Working Group
>attempting such a thing if you paid me! Meanwhile, as you say, ecommerce
>will surely happen regardless. So I'm not spreading doom and gloom; just
>claiming that reference is the weak spot when we come to formalise the
>'semantic' web.

Graham Klyne

Received on Tuesday, 4 September 2001 14:49:40 UTC