Folks, (01)
Before commenting on the responses to this thread, I'd
like to quote a comment that Cory Casanave made on
a different thread: (02)
CC> A valuable task for this group would be to collect and
> validate user driven requirements as well as the scenario
> of applying an upper ontology to those solutions. This
> will help nail down the set of problems we are addressing
> and serve as a set of test cases for the solutions to be
> applied. Such examples should be expressed in general but
> backed up with user driven and specific test cases. I
> suspect this kind of focus is the only thing that can
> resolve the theoretical debates. (03)
Absolutely! A representative set of test cases would be an
enormous aid to focusing these discussions. (04)
In fact, Cyc has addressed far more cases than any other
ontology on the planet. Although I have criticized Cyc many
times (largely because it's such a large target), if I were
forced to choose one single ontology today, the only one I
would seriously consider is Cyc -- primarily because it has
been tried and tested on a large and varied number of cases. (05)
On the other hand, if I did adopt Cyc, my first step would be
to modularize it from top to bottom. I would keep the entire
existing Cyc ontology as an option, but I would move nearly
all the axioms to microtheories -- including the ones at the
very top -- and make it possible to test alternative top
levels (or even no top level at all). (06)
To return to this thread, I'll start with a comment by Dale: (07)
DL> The word "lattice" connotes, however, an overly rigid
> structure that is not amenable to most "ontologists."
> Isn't it more a "mesh"? (08)
The lattice of theories comes for free when you adopt first
order logic or many subsets or extensions thereof. Nearly 70
years ago, Adolf Lindenbaum proved a theorem, which I discussed
in my theories.htm paper, that the set of all theories that can
be expressed in such a logic forms a lattice. (09)
So I would answer that a lattice is not "overly" rigid. It
is the natural "mesh" that results from your choice of logic.
You can choose to ignore it, but I recommend that we should
take advantage of it as a basis for relating and organizing
the modules (or microtheories) of an ontology. (010)
Some related comments by Dale and Nicolas: (011)
NR> You're convinced that the "one-size-fits-all" upper
> ontology is doomed for failure. But failure for what kind
> of problem? (012)
DL> Are you implying that the "one-size-fits-all ontology"
> of the contemporary physicist looking for the "theory of
> everything" (TOE) is a "disaster"? Do they know that? (013)
To answer Dale's last question: Of course they do. That's
why they would never accept any proposal as a final TOE
without an enormous amount of testing -- even then, they
would continue probing for flaws for centuries. They
certainly would not accept anything as a TOE just because
some TOE Working Group recommended it. (014)
As an example of a disastrous assumption that does not belong
at the upper levels, the philosopher Peter Strawson proposed
that "objects" should have "ontological priority" over processes
because processes cannot be identified without first identifying
the objects that participate in them. Nicholas Rescher ridiculed
that claim by saying that numbers should be given ontological
priority over people because people are typically identified by
social security numbers, employee numbers, etc. Alonzo Church
made an amusing response to a similar claim: (015)
http://www.jfsowa.com/ontology/church.htm (016)
But more important than the amusement is the fact that there
are no such things as primitive objects in physics. Even though
physicists don't yet have a TOE, all available evidence points
to the fact that processes are far more fundamental than objects
at every level. Even at the macro level, living organisms are
processes that cannot survive for even a short time without
continuous input of air, water, food, etc. (017)
This point, by the way, is one of the reasons why Cyc's upper
level is so complicated -- they're trying to accommodate both
views at the same time. That's one reason why I would prefer
to keep modules for birth & death, creation & destruction, etc.,
out of the upper levels and only bring them in if the current
topic requires them. (018)
Barry made several detailed comments on my note: (019)
BS> Actually I think BFO, SUO and DOLCE agree on very much; an
> active effort to merge BFO and DOLCE is under way, and I plan
> to attempt to initiate a similar effort with SUO in the future. (020)
That's good progress. But whenever ontology is being discussed,
the big elephant in the room is Cyc. As I mentioned above, Cyc
has addressed a far greater range of issues than any other proposed
ontology. I would have much more confidence in a merger that
included Cyc than one that ignored Cyc. (021)
JS>> Yet people have been communicating successfully for thousands
>> of years with very few common assumptions about top-level
>> entities, such as time, place, object, process, etc. (022)
BS> There are philosophers, it is true, who have very strange
> assumptions about some of these things; but common people share
> very many of these assumptions; each that there is an earlier
> and later, that some objects are closer together than others;
> that objects can undergo processes of change; that objects can
> be destroyed, etc. (023)
I'm all in favor of having modules for time, space, etc., in the
ontology, but only as replaceable options for a 3-D vs. 4-D view,
a Newtonian vs. relativistic time or even for a nonmetric time
(in cases where the causal ordering is fixed, but the durations
may be unpredictable and difficult or impossible to measure). (024)
BS> There are now beginning to be examples of cases where strong
> ontologies were able to resolve some of these problems. (025)
Those would be excellent use cases for testing and evaluating
ontologies. But I would also like to ask what use was made
of the axioms, what kinds of axioms were used, and what level
of detail was needed for different applications. (026)
JS>> The most successful sharing in *all* fields -- science,
>> engineering, medicine, business, etc. -- has been based on
>> *terminology* at lower levels with very few, if any axioms
>> about the upper levels. (027)
BS> I think this is just wrong. When scientists (to take just one
> example) use variables (x, y, z, t, etc.) to formulate differential
> equations there is a complex web of axioms underlying such use. (028)
There are very few unarticulated assumptions at the fundamental
levels of physics other than the assumption of regularity in
nature and the belief that there exists something outside our
minds. Over the past few centuries, physicists have invented
numerous mathematical formalisms with widely varying structures,
but equivalent predictions. (029)
For example, the global _principle of least action_ (which in
modern terms is represented by the Lagrangian) is formally
equivalent to the localized differential equations. In quantum
mechanics, Heisenberg developed matrix mechanics and Schrödinger
developed wave mechanics from very different assumptions, but they
were again proved to be formally equivalent. For these issues,
I highly recommend _The Road to Reality_ by Roger Penrose. (030)
BS> These axioms are, it is true, rarely explicitly formulated. But
> that is in part because they are mostly trivial; in part because
> scientists themselves are tacitly perfectly familiar with them.
> when scientists from different disciplines need to interact, then
> some of these axioms do become explicitly formulated (as, again,
> in the sphere of biomedicine). (031)
I agree that everyone in every endeavor has an enormous number
of unarticulated assumptions. And Cyc is the group that has done
more than any other to recognize them and formalize them -- that
reinforces my earlier point about Cyc. (032)
JS>> For reasoning and computation, the axioms should be introduced
>> at the lower, problem-oriented levels. (033)
BS> The current movement in biomedical informatics (which is the
> area I know best) points in an exactly opposite direction. (034)
Perhaps, but I'd like to see the use cases. (035)
JS>> In short, the hope of finding a detailed common set of axioms
>> at the upper levels is *DOOMED*. On the other hand, a very
>> simple upper level with very little detail would be possible. (036)
BS> As history shows, scientific nihilism is usually a bad idea... (037)
Yes, but I'm recommending pluralism (an infinite lattice), not
nihilism. However, for most of these cases, we are debating
the boundary between detailed and simple. That is best decided
by studying the use cases and by including the group with the
most experience -- Cyc. (See also the discussion below about
situation calculus vs. pi calculus.) (038)
JS>> For example, the upper level might say that there exist such
>> things as objects and processes, but not make *any* distinction
>> between the two. (039)
BS> Soup, eh? (040)
On the contrary, what I call the knowledge soup results from
a disorganized accumulation of too much detail. The process
of "crystallizing theories out of knowledge soup" consists of
selecting the minimum axioms necessary for solving a problem. (041)
Some responses to Chris: (042)
CM> Surely it is important that researchers within that very
> large community mean the same things by such terms as "blood",
> "gene", "ribosome", "protein synthesis", "cellular degeneration",
> etc, and that, in particular, they can be sure that the results
> of any reasoning upon those terms can be shared consistently
> within the community. (043)
Certainly. But I would hardly consider those categories to be
in the upper-level ontology. And when you start talking about
research on protein synthesis, any upper-level assumptions
about distinctions between objects and processes are almost
certainly going to be inconsistent with the physics at the
molecular level. (044)
CM> Generalizing the point to the thousands of salient terms within
> the biomedical community, a fairly elaborate common upper-level
> ontology would seem to be essential if sharing is to be possible
> within that community at all. (045)
There is a very big difference between sharing data and accepting
conclusions. Sherlock Holmes, for example, would never accept a
conclusion from Scotland Yard without seeing all the original data,
the assumptions made, and the steps of the reasoning. The same is
true for professionals in any field -- ranging from medicine and
law to Egyptian archeology. (046)
Any complex conclusion that is being shared should also specify
exactly which sets of axioms were used to derive it. That is why
I recommend the metadata registry for cataloging those modules in
the lattice of theories that are frequently used and reused. (047)
Nicolas asked a large number of very pertinent questions: (048)
NR> what kind of meta-ontology of problem solving do we need to
> organize/classify/recognize/discriminate ONTAC problems? (049)
This and the other questions in your note are very important.
I don't know the answers, but I believe that this group must
address them before making any firm recommendations. (050)
Leo made the following point: (051)
LO> I draw different conclusions from the experience of SUO,
> and largely programmatic rather than substantive in nature. (052)
I'm not sure where we differ, because I agree with the points
you made following this remark. (053)
The following remark by Nicolas shows why the detailed axioms
for time should be kept out of the upper levels: (054)
NR> Besides PSL and FLOWS having a solid formalization in CL or
> a dialect of CL (e.g., a KIF-like dialect for PSL) both PSL
> and FLOWS are ontologies focused on the notion of "process"
> and associated concepts, e.g., object, activity, activity
> occurrence, etc... (055)
A major problem with PSL is that it is based on the situation
calculus, which has been widely used in AI, but by practically
nobody outside of AI. In fact, three major areas use pi
calculus rather than situation calculus: (056)
1. It's implemented in the Erlang language, which is the
primary language for programming telephone networks.
(Robin Milner, who developed the pi-calculus, had been
a consultant to both ATT and British Telephone.) (057)
2. It's implemented in LOTOS, which has become an ISO
standard, and which is used very widely by NASA for
the very complex and delicate temporal reasoning
problems in their space missions. (058)
3. It's being used as a foundation for the methodologies
for Business Process Modeling. See, for example, the
book by Michael Havey, _Essentials of Business Process
Modeling_, which contains a good overview of the pi
calculus and how it is being used in practice. (059)
When I talk about disasters, I mean the possibility of
requiring situation calculus in preference to pi calculus.
Forcing that assumption would create inconsistencies with
the telephone industry, the aerospace industry, and
business data processing. (060)
I would have no objections to including modules for both
situation calculus and pi calculus in the ontology, but I
would strongly object to making either one a requirement. (061)
John Sowa (062)
_________________________________________________________________
Message Archives: http://colab.cim3.net/forum/ontac-forum/
To Post: mailto:ontac-forum@xxxxxxxxxxxxxx
Subscribe/Unsubscribe/Config:
http://colab.cim3.net/mailman/listinfo/ontac-forum/
Shared Files: http://colab.cim3.net/file/work/SICoP/ontac/
Community Wiki:
http://colab.cim3.net/cgi-bin/wiki.pl?SICoP/OntologyTaxonomyCoordinatingWG (063)
|