[Top] [All Lists]

Re: [ontolog-forum] standard ontology

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: "John F. Sowa" <sowa@xxxxxxxxxxx>
Date: Sat, 21 Feb 2009 11:33:34 -0500
Message-id: <49A02CDE.90602@xxxxxxxxxxx>
Don and Pat C,    (01)

Before getting to Pat's comments, I'll start with Don's, which
I strongly endorse.    (02)

DC> My experiences with DOD affirm your [Pat C's] comment in that
 > they want an order of magnitude advance from semantic technologies.
 > But the hard truth is that the more tightly a domain is focused,
 > the more a relational DB solution will equal, or better a semantic
 > solution.  At much less cost and difficulty.    (03)

The only point I disagree with is the implication that a relational
database is *not* a vital part of a semantic solution.  I realize
that the current SQL standards leave a great deal to be desired,
and most of the limitations were forced on SQL by the awkward
initial design, which IBM considered an unfinished research project.    (04)

Other relational DBs (the original Ingres, for example) had much
better query languages.  Unfortunately, Oracle took the raw SQL
from the IBM research reports, commercialized it, and made it
a de facto standard.    (05)

When Ted Codd saw Prolog, his initial reaction was "I wish I had
invented that."  The Datalog language is essentially a very clean
subset of Prolog designed for DB assertions, constraints, rules,
and queries.  A typed version of Datalog would be an excellent
upgrade from SQL *and* the Semantic Web.    (06)

As evidence for the thirty years of outstanding R & D on semantics
related to databases (both relational and object-oriented), I'll
cite the series of books on DB semantics:    (07)

    http://www.informatik.uni-trier.de/~ley/db/conf/ds/index.html    (08)

This is a series of nine books, which are based on conferences
sponsored by the IFIP Working Group TC2/2.6 on Database Semantics.
The first one was in 1985, after 15 years of R & D on both the
practical and theoretical issues of relational databases.  There
are many other excellent books on deductive databases and their
relationship to knowledge bases, but these books include a good
sampling of that work.    (09)

The last book in this series appeared in April 2001.  Not so
coincidentally, there was a major gathering of the early
Semantic Web tribe at Stanford in July 2001.  Somebody had
said that the Semantic Web "sucked the wind out of" all those
other projects, but Pat Hayes defended the Semantic Webbers
by saying that there were many outstanding people working
on SW projects.  I agree.  Those are people who were sucked
out of other semantic projects by the multimillions of euros
that were dumped on the SW.    (010)

DC> Fielding large scale, operational semantic systems is not
 > yet feasible because the supporting infrastructure is not
 > in place. For example, look in the Sunday paper job ads for
 > oracle DB administrators.  Then look for ontologist ads...    (011)

Nearly every major commercial web site (and many, many minor
ones) are built around relational databases.  If the SW
designers had chosen Datalog as their primary language and
integrated it with both relational and network DBs, they
would have built the supporting infrastructure long ago.
The DB administrators and the ontologists would be working
together seamlessly, and the number of job ads for the
combination would be far greater than the sum for both.    (012)

DC> Unfortunately, as it takes money to make money, success
 > breeds success. Fielded semantic systems that work will
 > garner more funds for the application of those technologies.    (013)

I agree.  And there are many small (and some very large) groups
that are combining logic-based languages such as Prolog, Datalog,
and others with very large relational and object-oriented DBs.
An example is Experian -- one of the three large companies that
monitor everybody's credit.  They not only use Prolog for their
inferences, they bought the Prologia company, which was founded
by Alain Colmerauer, the person who implemented the first version
of Prolog.  If you type "experian prolog" to Google, you'll get
5,870 hits.    (014)

You won't see many published papers about such applications
because (a) they're no longer research issues, (b) the people
who are using Prolog are busy working on practical problems,
and (c) the successful companies don't want to tell their
competitors how they make money.    (015)

DC> The DOD entities I've dealt with typically seek to integrate
 > some number of relational DB's, sometimes with unstructured data.
 > But within a fairly tight domain (yes, there is a range to the
 > domain focus).
 > Interoperability, while given much lip service, lags the commercial
 > world. The DOD XML Registry was going to be the solution...still
 > waiting on that one. By the way, where is the ontology repository
 > and how capable is it?    (016)

I like the idea of a registry/repository, but there is a serious
issue with any open source project:  giving things away for free is
usually a poor business model.  Eclipse ( http://www.eclipse.org/ )
is an outstanding example of a successful project.  But it was
started by IBM as a proprietary platform for software development.
When IBM realized that they couldn't make a profit from Eclipse,
they gave it away to an industry consortium.    (017)

Linux is another example of a successful project that began with
a well-defined target (Unix), a large collection of tools (Gnu),
and a benevolent, dedicated dictator who also happened to be
a superprogrammer (Linus Torvalds).    (018)

OpenOffice is an outstanding freebie, which also began as a commercial
system (StarOffice), which Sun bought and couldn't sell profitably.
So they gave away the code and sell services based on the StarOffice
superset.  But Sun has not been successful in attracting many
free contributors to the OpenOffice code base.    (019)

I think that the Eclipse model is the best one to emulate, but that
means the emphasis would have to be on the development tools, which
have to be very good in order to attract a significant number of
users and contributors.    (020)

PC> ... we are not starting from scratch, but using *everything*
 > freely usable from the OpenCyc???    (021)

I wasn't looking at OpenCyc, but at Cyc, which is the closest thing
to what you're proposing.  But Cyc hasn't been able to solve any
of the interoperability problems well enough to provide a positive
return on investment.    (022)

PC> It seems you haven't read my point-by-point enumeration of
 > the reasons.    (023)

I've read them all.  None of them answer the question why Cyc has
not already solved the interoperability problem -- even partially.    (024)

PC> I have already detailed reasons why Cyc in its present form and
 > with its present management cannot serve as a common foundation
 > ontology acceptable to a broad community.    (025)

The present management and stockholders would be delighted to be
bought out for a fraction of the funding you're asking for.    (026)

PC> Where Doug's original model failed is that he thought that he
 > only had to develop a competent ontology, and could then license
 > it to users....    (027)

No.  Doug and his group worked long and hard with very competent
people at universities, government agencies, and stakeholders at
some large companies with very good research divisions.  Those
companies had large applications, and they had top-notch people
with excellent backgrounds in AI and comp. sci.  The license they
signed with Cycorp gave them *all* Cyc software with unlimited,
royalty-free rights to use in any way they saw fit.  But none
of them found any useful applications for Cyc.    (028)

PC> ... to develop a community that uses some FO that can translate
 > from Cyc and any other ontology into each other, and to demonstrate
 > its usefulness by producing open-source applications and utilities,
 > especially an NL interface, to make it easier to use.    (029)

That is the wishfulest of wishful thinking.  Which open-source model
are you proposing?  Eclipse?  Linux?  OpenOffice?    (030)

All three of them started with a very precisely defined structure.
In comparison, FO is pie in the sky with a vague hope that somebody
someday might discover something useful.    (031)

PC> I have said repeatedly that if there were a broadly representative
 > community of users that openly shared their applications and
 > utilities, any competent foundation ontology, including Cyc,
 > could be used as a starting point...    (032)

Where are you going to find that Utopian community?  Please look
at the existing open source projects.  The overwhelming majority of
projects hosted by Sourceforge are dead or moribund.  Competitors
don't share their applications and utilities because they want
to make money from them.  The only ones they share are ones they
can't make money from -- e.g. IBM's contribution of Eclipse, Sun's
contribution of OpenOffice, or Cyc's contribution of OpenCyc.    (033)

PC> Inertia is immense.  Getting someone to put up money to prove
 > that the ontology will help their applications is very hard.    (034)

I talked with very knowledgeable managers at research groups that
had paid to be Cyc stakeholders.  I asked them how they were using it.
They said that people in their departments had tried to use Cyc for
various applications, but none of them could demonstrate that it gave
them any advantage.  One manager said flatly, "Several people have
worked with Cyc over the past few years, and all of them were fired.
I don't believe that was a coincidence."    (035)

PC> I feel certain that it is impossible for anyone person to create
 > a demo that is sufficiently more impressive than traditional IT
 > methods so as to break through the consciousness of managers who
 > could fund real projects.    (036)

The manager I mentioned above had an excellent knowledge of both AI
and traditional methods of software development.  The people who were
fired were PhDs who were given the chance to collaborate with IT
people in their company on those projects.  But they *failed*.    (037)

PC> As an example, consider language understanding....    (038)

At VivoMind, we got very good results in language understanding,
and we don't use anything that remotely resembles your FO.  See
the following talk and the URLs in the last slide:    (039)

    http://www.jfsowa.com/talks/pursue.pdf    (040)

PC> The point I will emphasize, every time I hear such an argument,
 > is that the cost of *not* undertaking such a program (or multiple
 > such programs, if any others are proposed) is even higher.  That's
 > important enough to repeat.    (041)

Without evidence, that claim is wishful thinking, and repeating it
doesn't make it more convincing.    (042)

John    (043)

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
To Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (044)

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