[Top] [All Lists]

Re: [ontolog-forum] Truth (Tim B-L's vision for the SW in 2000)

To: ontolog-forum@xxxxxxxxxxxxxxxx
From: John F Sowa <sowa@xxxxxxxxxxx>
Date: Tue, 10 Jul 2012 22:27:47 -0400
Message-id: <4FFCE4A3.3010209@xxxxxxxxxxx>
William,    (01)

Your question can't be answered in one or two sentences.    (02)

> My initial impression was caused by a revulsion to the underlying
> assumptions [in Boris Motik's paper].  But I admit that it's not
> a bad paper of its kind. [For anyone who likes that kind of thing.]    (03)

> I ask if you have anything further to say about these "underlying
> assumptions".  They may go much deeper than the assumption that
> logics used by the semantic web should be decidable, to the
> *reasons* someone would have such a belief, i.e., the purpose
> of these logics and their languages.    (04)

For a brief history of semantic networks, see    (05)

    http://www.jfsowa.com/pubs/sement.htm    (06)

Early networks didn't represent logic very well.  Bill Woods (1975)
wrote a critical article called "What's in a link?"  His student,
Ron Brachman (1979), responded by implementing the Description Logic
KL-ONE for his PhD dissertation.    (07)

Brachman, Fikes, and Levesque (1983) combined KL-ONE with a first-order
theorem prover in a hybrid system they called KRYPTON.  They called
the KL-ONE hierarchy and and its classifier the T-Box (terminology box)
and the FOL statements and theorem prover the A-Box (assertion box).    (08)

The T-Box and A-Box combination was widely adopted in many systems.
The logic for the A-Box is usually much more expressive than the
logic for the T-Box.  For certain kinds of problems, a theorem prover
for a typed version of FOL that uses a T-Box with hierarchy of types
or sorts can run orders of magnitude faster than a theorem prover
for an untyped version of FOL.    (09)

During the 1980s, the DLs weren't very efficient as classifiers,
and two schools of thought developed about how to improve efficiency:
limit the expressive power of the T-box to a version of logic that
could be computed more efficiently; or use hybrid methods that
support different reasoning methods for different problems.    (010)

Cyc is the world's largest formal ontology and reasoning system.
Instead of just two components (T-Box and A-Box) it has multiple
reasoning methods specialized for a wide variety of problems.
For any particular problem, the system chooses an appropriate
method automatically.  See    (011)

    Fads and fallacies about logic    (012)

The ability to let the system choose the reasoning method
automatically can be, for many applications, more powerful
and efficient than just relying on a single method.    (013)

> While thinking that decidability is essential might not *itself*
> be justly called a "disease", only wrong, there may also be
> a more deep rooted notion about the relationship between people
> and machines that would lead to thinking decidability was important,
> and this fundamental notion that may be finding itself in conflict
> with what is most practical.    (014)

The simplest explanation is "To a man whose only tool is a hammer,
all the world is a nail."  Computational complexity is a subject
that every computer scientist should know, at least at a basic level.
But there are some people for whom decidability is their only tool.
They repeated the decidability mantra until people who had no idea
what it means began to believe it.    (015)

Following is a quotation from Bob MacGregor (1991), who developed
the LOOM and PowerLoom systems.  He used the T-Box and A-Box
combination and advocated a variety of methods:    (016)

> There is a variety of evidence suggesting that a KRS (Knowledge
> Representation System) with an expressive terminological language
> (and an incomplete classifier) is a more useful tool than a KRS
> with a small terminological language (and a complete classifier).
> In this section, we present arguments in favor of implementing
> an expressive ("promiscuous") terminological language, and also
> propose an architecture that may turn out to allow us the best
> of both worlds.    (017)

A more common term today is 'heterogeneous' instead of 'promiscuous'.
That is what Marvin Minsky has been recommending for years, and the
evidence I've seen has convinced me that he's right.    (018)

I have no quarrel with allowing people to choose a tightly constrained
language such as OWL, if they find it useful for their problem.  But
the heterogeneous systems are much more flexible.  That is what Tim B-L
proposed in 2000:    (019)

Tim B-L:
> The goal of interoperability between heterogeneous components that we
> build is one that will test the extent to which the Semantic Web is
> achieving its promise. The more diverse the systems interoperating,
> the greater the merit of the Semantic Web.    (020)

See http://www.w3.org/2000/01/sw/DevelopmentProposal    (021)

I never liked the use of XML for representing triples, but Tim also
mentioned a wide range of other languages that could be supported:
"Prolog, SHOE, Algernon and KIF".  He mentioned triples, but he also
mentioned relational databases.    (022)

If Tim's view in 2000 had been accepted, I think that we could have
had a great Semantic Web today.  It could have become a powerful
platform for developing logic-based applications.  Instead, all
that's left is an anemic combination of RDF(S) + OWL + SPARQL.    (023)

John    (024)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/  
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/ 
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J    (025)

<Prev in Thread] Current Thread [Next in Thread>