Re: LANG: need to CLOSE Issue 5.6 Imports as magic syntax

Jim,

I think that the solution that I proposed in [1] (based on ideas from
Mike Smith and Peter) is one that is reaching consensus among those
people who are in favor of imports. Note that this solution does not
mention trust or belief anywhere, it is soley based on deterministically
selecting a set of statements (from possibly distributed sources) and
computing the entailments of them. Furthermore, it does not constrain us
in later having more fine-grained mechanisms for trust and belief. Such
mechanisms could later decide whether to trust an imports, or to only
trust them partially, etc. Also, the "meat-cleaver" approach is the one
that has been used in ontology design for at least a decade and the
semantics I've proposed is similar in nature to the ontology inclusion
semantics of Ontolingua. Therefore, it seems that this approach is one
that would be useful to a large communtiy even if a fine-grained
approach was added later.

If we postpone this issue, then there is no way to commit to an
ontology. If there is no way to commit to an ontology, then there is no
point in having a semantics for an ontology, because the entailments of
documents do not depend on ontologies. Each document's entailments can
only be computed independently of all other documents. Sure, anybody can
select a set of documents, merge them, and apply the semantics, but this
is no basis for shared meaning because how can you guarantee that
somebody else will select the same set of documents? The only way is to
tell them in some method outside of RDF/OWL. But if we do that, then
this isn't much of an advance over XML Schema because in order to get
semantic interoperability, two humans still have to communicate using an
alternate mechanism.

If we postpone this issue then we admit defeat on some of our most
important requirements. We would have to say that we cannot accomplish
design goal 3.1 Shared ontologies (note this goal specifically says RDF
alone is not sufficient), and requirements R3 (Explicit ontology
extension) and R4 (Commitment to ontologies). I believe that the user
community was impressed with our requirements document, and thought that
as a result of it we were committed to considering the tough issues that
would make the Semantic Web work. If we demote these requirements (which
would be necessary if we postpone the issue), then our working group
loses serious credibility. This is not a decision to take lightly.

Considering that we have a solution that addresses the requirements,
that simply clarifies things from DAML+OIL, that is a common element of
ontology design, and that does not restrict us from developing
alternative approaches in the future, I can see no good reason to
postpone the issue. I think that anyone who opposes this solution should
come up with an example of an application that it breaks or a specific
language extension that it will prohibit.

Jeff

[1] http://lists.w3.org/Archives/Public/www-webont-wg/2002Nov/0004.html

Jim Hendler wrote:
> 
> *** Chair Hat Off ***
> 
> I've spent a lot of time this week grappling with the issue of
> imports, talked to a lot of people, and reread all the emails and a
> number of discussions on RDFIG.  Based on all this I'm reluctantly
> forced to concur with Dan that the right thing for us to do is to
> POSTPONE this issue.  The rest of this message tries to explain my
> opinion.
> 
> ----
> To start with, lets create a very simple ontology I need for an
> example (I'll use N3 to save space):
> 
> <> a owl:ontology.
> :animal a owl:class.
> :person a owl:class.
> :dog owl:sunClassOf :animal.
> :owns a rdfs:property;
>    rdfs:domain :person;
>    rdfs:range :animal.
> :Dan :owns :dog.
> 
> So if you read this document and I ask you "Do you believe there is
> someone who owns an animal" then if you trust the document, you
> should agree with this statement.
> 
> ---
> case 1
>   Suppose we now make a slight change to the ontology.  There is a
> public file with information about people who work at the W3C (the
> actual file is at [1]) which we will call W3Cfile: for short.  So
> instead of
> 
> :Dan :owns :dog.
> I replace it with
> W3Cfile:Connolly :owns :dog.         # skip some details of fixing
> the namespace
> 
> now, I ask you "Do you believe someone from the W3C owns an animal".
> Seems to me to say Yes to this, you need to believe my document and
> also to believe W3Cfile.  If you don't believe W3Cfile,  but you do
> believe my document, you would say "I believe someone owns an animal,
> but I'm not yet convinced this is someone from the W3C."
> 
> -----
> case 2
>   Jeff makes an extremely important point - suppose we do want to make
> it so that you should believe someone from the W3C owns a dog just by
> believing my document -- then if I include in my document
> 
> <> owl:imports W3Cfile:.
> 
> (and we do something similar to what Peter suggests) it would be a
> way of saying my document trusts everything in the W3Cfile, and
> therefore just by trusting me, you are convinced to believe someone
> from the W3C owns an animal.
> 
> Jeff is absolutely correct that this is an important feature of the
> Semantic Web.
> 
> --
> However, here's the problem -- notice that I had to discuss this
> using terms like "believe" and "trust".  Further, there's a lot of
> other issues as well.  Consider, that I might want to say "I trust
> the W3C for issues relating to Dan Connolly, but not for anyone else"
> or "I don't trust how the W3C discusses animals, so ignore what they
> say about that and use my definitions," or... or ... or...
> 
> Putting words into Dan Connolly's mouth, he is worried that even
> though Jeff's concern might be legitimate, we run the risk of taking
> a meat cleaver approach to something that could be better handled
> with a scalpel -- by accepting this approach, we may discover later
> that we regret it, but lots of implementations would already relay on
> it - whereas if we focus on this issue in our research and
> implementations, at a later time we may be able to do a much more
> comprehensive and better solution and then make that normative.
> Also, it may be important to do this at the basic document level
> (i.e. in some RDF way) instead of in Owl per se.   I've come to agree
> with that, and therefore suggest POSTPONING.
> 
> ----
> btw, I have no objection to us doing something "non-normative" with
> respect to what Jeff and Peter have suggested.  I think if we put
> that into an Appendix (non-normative) or in a separate note (or a FAQ
> page) - I would not complain -- but making this definitive, I've come
> to believe Dan and his colleagues may be right.
>   -JH
> 
> [1] http://www.ece.umd.edu/~adityak/w3cpeople.rdf
> 
> --
> Professor James Hendler                           hendler@cs.umd.edu
> Director, Semantic Web and Agent Technologies     301-405-2696
> Maryland Information and Network Dynamics Lab.    301-405-6707 (Fax)
> Univ of Maryland, College Park, MD 20742          240-731-3822 (Cell)
> http://www.cs.umd.edu/users/hendler

Received on Monday, 4 November 2002 10:17:17 UTC