Re: SUO: Evaluation Framework for Content Standards
At 10:39 AM 6/16/2003 -0700, John F. Sowa wrote:
Nicola Guarino just posted a note
(copy below) about
a 13-page PDF document with the above title. He called it
a "first attempt", and as such, I believe it is something
that the SUO Working Group should consider. But it also
omits many important issues and considerations that have
been mentioned on SUO list. Its primary advantage, however,
is that it is a clear statement of a certain point of view,
and it can help to focus the debate.
I'd also like to comment on some related points that have
been raised on SUO list. Some comments by Erik Larson:
EL> My point is that building the ontology isn't the same as
> lexicographers developing their definitions. The
ontology
> has to take some non-trivial normative stance on the sorts
> of things there are in the world. That's the whole
point.
That point assumes that this committee is omniscient, or at
least, that we are better qualified to tell other people how
to mind their own business than they are. I know that I am
not omniscient. And I have no confidence whatever in anyone
else's qualifications to take "some nontrivial normative
stance on the sorts of things there are in the world."
Every attempt that anyone has proposed so far has been a
disaster, and I have never seen any suggestions that have
the slightest chance of getting anything better.
EL> The reductio to making subtypes of every type whenever
> someone has a new interpretation is that eventualy the same
> complexity that the ontology was supposed to help ameliorate
> will be reproduced within it.
Of course. But there is no other way short of becoming God
or a fascist dictator. The first is unlikely, and the second
would be a disaster.
EL> But I start to get off the boat if the idea is that we'll
> never have to make decisions about the sorts of entitities we
> want in an official SUO. We'll have to make those
decisions.
> Surely you can agree with that.
I believe that we will have to make many decisions, primarily
on the choice of distinctions that can best accommodate every
conceptualization of the world that anyone might require for
any purpose whatever. Those choices are empirical issues that
depend on examining the evidence. A methodology, such as
formal
concept analysis (FCA), can help by showing the consequences
of any decision. And FCA is one of the formalisms that are
packaged in the IF framework.
Mike Pool also commented on my earlier note:
> In this case, much as I hate to do it, I'll push the
> Jean-Luc Delatre line and emphasize how dangerous it is
> to assume that ontology merging is merely a case of mapping
> terms to one single generic upper ontology.
I'm not sure what you mean. If you mean that you have
finally come to the realization that JLD is correct, then
I am delighted to hear that -- because I also believe that,
with minor qualifications, his view that no single ontology
has ever been adequate or is ever likely to be adequate must
be the starting assumption for any serious work on ontology.
From the acknowledgement, that I've always been happy to make, that it
is hard to merge ontologies, it does not follow that one denies that a
single ontology can be adequate. It just acknowledges that some
people will have a heck of a time making their ontology compliant with
the standard if a single one is adopted. I think that I part
ways with Jean-Luc, it's hard for me to figure out where I part ways with
you, because it has never been my goal to use the SUO alone to solve
interoperability problems. Once we give up that objective,
having a useful standard upper ontology (one that does not seek to
represent every single possible but incompatible sense of 'event'
'process' and the like) becomes much more reasonable.
I am willing to admit the
hypothesis that JLD may be wrong,
but the overwhelming weight of evidence is on his side.
Anyone who claims otherwise either doesn't understand the
problem or is a starry-eyed optimist.
However, I believe that your next sentence misses the mark:
> Your glib trivialization of the problem surprises me.
The problem that we were discussing concerned resolving
differences between Cyc #$Event and SUMO &%Process (check the subject
line under which you sent your note) and the "solution" to
which I was responding is the following:
- To take the example of "event", there is no need to
- force a decision. The ontology can simply include
- three types: Event4D, Event3D, and a generic Event,
- which includes Event4D and Event3D as subtypes.
This purported solution puzzled me as it wasn't at all clear how it
served to solve anything close to the problem under discussion.
Following is a restatement of a
non-trivial method that many
people have successfully used for accommodating multiple
views within a single ontology:
1. When you have two different axiomatizations of a single
concept, such as the word "event" in
English, you can
introduce them as subtypes of a single type called
Event.
2. Given two axiomatizations, such as a 3D and a 4D view of
events, rename one Event3D and the other event
4D.
3. Then take the common intersection of the axioms for both
Event3D and Event4D and associate them with the type
Event.
4. Finally, use the type Event as the default target for
mappings to and from the English word
"event".
5. When more information from the context becomes available,
the mapping can be specialized to whichever of the
subtypes
is most appropriate.
By using this method of renaming concept types, it is possible
to accommodate all the versions within a single ontology. But
as Mike said, it may make the ontology very complex.
The alternative is to have a modular ontology. SUMO has 11
modules, which the developers believe are all consistent.
The Cyc developers are more realistic. They have 6000
microtheories, and they are intended to accommodate
inconsistent viewpoints. My major reservation about Cyc
is that they do not go far enough in organizing their
microtheories and in pushing the options to the higher
levels of their ontology.
Summary: Those people who want a single ontology can have
one by adding multiple nodes for every specialized sense.
I prefer to have a modular ontology with alternatives. The
question of how many nodes or how many modules are needed is
an empirical issue that cannot be resolved by taking a vote.
Following is the note from Nicola (with the URL for his paper).
John Sowa
______________________________________________________________
Dear all,
this is to communicate that a new deliverable has
recently be uploaded on the OntoWeb site: Delieverable 3.5 - Evaluation
Framework for Content Standards. It elaborates on contributions presented
at OntoWeb SIG 1 meeting held in Innsbruck in December 2002. Among other
things, it presents a first attempt at characterizing the notion of
"ontology quality* in a rigorous logical way. The paper is
downloadable at
http://ontoweb.aifb.uni-karlsruhe.de/Members/ruben/Deliverable%203.5
Comments welcome.
Best,
Nicola
--
Nicola Guarino
Laboratory for Applied Ontology (LOA), ISTC-CNR
Institute for Cognitive Sciences and Technologies
National Research Council
Via Solteri, 38
I-38100 Trento
phone: +39 0461 402088
fax: +39 0461/435344
email: Nicola.Guarino@ladseb.pd.cnr.it
http://ontology.ip.rm.cnr.it
Provisional mirror site:
http://www.isib.cnr.it/infor/ontology/ontology.html
_______________________________________________
seweb-list mailing list
seweb-list@informatik.uibk.ac.at
http://informatik.uibk.ac.at/mailman/listinfo/seweb-list