Re: help wanted: RDF issue rdfms-assertion

patrick hayes wrote:
> 
> Jeff Heflin wrote:
> >
> >I think bringing lawyers into the matter only muddies the waters. What
> >is important is as Tim said "we must define the core algorithms for
> >determining meaning without hesitation or ambiguity." When someone
> >publishes something on the Web, they should be able to determine exactly
> >what is implied by their statements, and others should be able to
> >determine this as well.
> 
> Sure, but that isn't the main issue here. What about what is implied
> by their statements *together with other statements made by others* ?
> In RDF it is relatively easy to trace back any conclusions to their
> source, since entailment is so simple; but even in RDFS, a subclass
> relation might be inferred from two others published by different
> people, and the conclusion not be derivable from either one of them
> alone. So in general it is *impossible* to determine all the
> conclusions that anyone might draw from statements you make, if they
> are also allowed to use other statements. And RDF has been designed
> under the assumption that they are free to draw conclusions from
> *any* published content.

I actually think both are important issues. As you say, when someone's
statements are combined arbitrarily with someone else's then there is no
predicting what might be entailed, because we can't predict what someone
else might say. Therefore I think its important that it be possible for
reasoners to execute in a restricted mode that only uses statements from
a particular source and whatever ontologies or other sources are
explicitly commited to by that source (note RDF does not allow a source
to commit to ontologies, but it is one of OWL's requirements). This will
in some sense give the asker an idea of what the author originally meant
to convey, without it being warped by extraneous information. I see this
particularly useful in e-commerce settings where it is important that
everybody knows the terms of the contracts they are agreeing to.

Of course, if you couldn't combine these statements with those made by
others, then the Semantic Web would be severely limited (it wouldn't be
very distributed). How to allow this combination in a useful and
repeatable way is a very difficult problem. I think when queries are
issued in such a mode, it is important that either 1) users have some
way to specify which statements from which sources are acceptable to use
or 2) the system produces a proof of the answers annotated with sources
for each statement used in the proof. Unless we have one of these, I
don't see how we could ever rely on any conclusions that might be
produced..

Jeff

Received on Tuesday, 11 June 2002 13:41:56 UTC