Re: SUO: Organizations/Positions
Doug,
At 08:44 AM 9/6/2001 -0700, Douglas McDavid wrote:
>Adam --
>
>Some further comments marked below as: *DM -
>
>
>Adam Pease <apease@ks.teknowledge.com> on 09/04/2001 11:19:33 AM
>
>To: Douglas McDavid/Boulder/IBM@IBMUS, Ian Niles <iniles@teknowledge.com>
>cc: "Standard-Upper-Ontology (E-mail)" <standard-upper-ontology@ieee.org>
>Subject: Re: SUO: Organizations/Positions
>
>
>
>*DM - big snip
>
> >
> >Now, it seems that you (Ian) are reaching a point of personal comfort,
> >where you are wanting to introduce a small set of axioms into the
>uppermost
> >realm of human knowledge and concern (a standard, upper-level ontology).
> > >From my point of view, what you are proposing is a thin, brittle set of
> >concepts from a subdomain of what I have been talking about here.
>
>I was with you until you described the ontology as "brittle". In a
>physical system, something brittle breaks into pieces easily when stress is
>applied. In an ontology, I would suggest that "stress" is the addition of
>a new example or category of knowledge. "Breaking" would mean that drastic
>reorganization of the ontology is required to accommodate the new
>requirement. Any ontology will be incomplete with regard to some area of
>knowledge so the question must be whether the ontology can be elaborated
>rather than changed and reorganized. Our defense against brittleness is
>the peer review process. If someone from the group comes up with a
>concrete example that can't be accommodated that would be an indication of
>brittleness.
>
>*DM - I intended to use the term brittle in the sense I remember Doug Lenat
>using when he talked about the purpose of Cyc, back in the mid-80's. This
>is in sense that knowledge from one domain (say medical diagnosis) breaks
>when applied to another domain (say auto mechanics). The example Doug gave
>was a car with rust spots that was diagnosed as having the measles. My
>point
>about the organization and position discussion is that it is likely to
>break
>when applied to human social systems that are not in the commercial domain
>that seems to be the focus of recent discussion.
It sounds like we understand "brittle" in the same way. However, unless
you can provide a concrete example of where the proposed scheme breaks, I
think it's unfair to call it brittle, just as it would be unfair for
someone to call perdurantism brittle without providing an example that
shows its brittleness.
>*DM - snip
>
> >On the basis of this, I'd like to raise the following questions for
> >discussion:
> >
> >1. What is the method by which we assign certain concerns to the SUO
>level
> >and others to the level of subtending domain ontologies?
>
>My preference would be for us to work on any reasonably general area of
>knowledge (although this is certainly a fuzzy judgement) and worry later
>about what to exclude. One approach as we reach a reasonably point of
>stability would be simply to take only the top few thousand concepts and
>consign the remainder to domain specific ontologies. The choice would be
>an empirical one. I feel that arguments a priori about what to include or
>exclude would be ungrounded and impossible to resolve.
>
>*DM - My reading of a lot of the discussion over past months is that you
>have
>been very consistent in theis position, and this is consistently the source
>of
>a lot of the angst expressed by others about the method, or lack of method,
>in the approach that is being taken with SUMO. I'm trying to sharpen this
>debate, because it is at the heart of the part of the PAR that I personally
>feel has the greatest hope of tractability in human lifetimes.
Chris Welty has proposed a method based on his recent AAAI tutorial which
may be a good candidate.
> >2. How do we propose, over time, to address whole sets of concerns such
>as
> >the ones I have raised here?
>
>I believe it would be for an individual or group to raise an issue about an
>area of knowledge, proceed to more formal statements in English about a set
>of concepts and then to propose formal axioms, consistent with SUMO or IFF
>(or some other future proposal).
>
>*DM - An alternative would be to consider a number of existing, and then
>subsequently proposed, ontologies for how they are or are not able to
>interface in an acceptable manner to the upper ontology. This would
>require a definition of acceptable interface, which I am more and more
>coming to believe is the crux of what might be possible to standardize.
I'm not sure what you mean exactly by 'interfacing' with an ontology. We
have, as you know, merged in some 40 different existing ontologies into
SUMO. However, I think we should be clear. The PAR says we're creating a
standard ontology, not a mechanism for creating ontologies. That might be
a worthwhile effort, but it would be a different effort.
> >3. Is it part of our methodology to reduce very large domains to a
> >minimalist set of concepts and axioms?
>
>I would hope so. Our set of axioms should be as simple as possible but no
>simpler. We can drive the development of axioms with examples. If
>expression of a concrete example cannot be done in a particular ontology,
>that should point to the change or elaboration that is needed.
>
>*DM - Which begs the question of what does it mean "to be done" in an
>ontology? My original note was prompted by the sense that some folks
>believed
>that having drawn this picture, with it's accompanying SUMO statements
> >
> > CognitiveAgent
> > |
> > |
> > OrganizationUnit
> > / \
> > / \
> > / \
> >Organization Position
>
>we had arrived at some point of "doneness" with respect to the subject of
>human organizations.
I would say merely that all the open modeling issues about organizations
that had been raised were addressed successfully in the proposed
model. Any ontology can be elaborated, just as any object model or
software system can be given new features. A common-sense judgement call
is made as to whether to issue a "release".
> A lot of what I have snipped out from my original
>note was to indicate how far I think this is from being done. "Doing" a
>concrete example, in my mind, means that you can not only make some
>statement that is uncontroversially true, but that you can fully explore
>the meaning that is necessary for the task at hand.
I agree. Examples can show where an ontology may be flawed but you can't
prove it's ok for all applications. This is similar to the debates
recounted in Loux's Metaphysics. A peer-review process tries to poke holes
in the theory. When those discussions tail off and no open problems exist,
that's a good time to accept the current model. Of course, if someone
later raises a specific problem, the debate should be reopened.
> So, to say that a
>Client Service Representative is a position and the Queen of England is
>a position, while maybe acceptably true, the really interesting stuff is
>all the dimensions along which these two things (and so many others) are
>different. For those of us thinking about the uppermost realm of ontology
>I think it is important to consider whether the uppermost ontology would
>ever have all the concepts that would articulate these interesting
>differences,
>or is there a principled manner in which multiple domain-specific
>ontologies
>would be brought to bear on such matters.
We might ask when should the Java SDK be called "done" - probably
never. But that doesn't mean of course that there aren't releases. Is
there some test for when the SDK is done? Well, there should be good
documentation, and no documented bugs. Similarly, the SUO shouldn't have
major open issues or documented logical contradictions when we enter the
draft standard phase. But determining when to stop will always be, in
part, a judgement call.
> >4. Do we want to subject each accepted set of terms and axioms to an
> >evaluation against the terms of the PAR? For instance should we expect to
> >be able to articulate how the set of terms below provides for:
> > a) a framework for domain ontologies,
> > b) interoperability of various applications,
> > c) mapping of data elements,
> > d) educational systems, and
> > e) disambiguation of natural language statements?
>
>I think this would be a reasonable exercise. In terms of the proposed
>formalization of organizations:
>
>a) the ontology could be expanded to handle legal frameworks or business
>rules of specific organizations
>
>*DM - Or, alternatively, the upper ontology could be refined to recognize
>that
>there are legal and business ontologies that need to be invoked, and the
>mechanism by which it is possible to determine when and how to do so.
yes, doing a domain specific ontology for legal reasoning that uses SUMO
would help to show whether SUMO itself has flaws or gaps.
>b) the ontology might support a set of back office IT systems that must
>communicate with each other about supervisory, legal and financial
>responsibilities within a complex company such as Shell.
>
>*DM - Or a set of ontologies, operating together under a standard upper
>mechanism might do so.
hmm, again, we're not developing a mechanism, but an ontology.
>c) the ontology could be "compiled" into a relational database that
>captures the terms and relations in the ontology, where the axioms serve as
>formal definitions for the database metadata
>
>*DM - Or a set of ontologies might be so compiled, with a compiled database
>version of some upper level as a means of navigating the domain-specific
>modules.
>
>d) the ontology could support an educational simulation that employs legal
>rules to advise a student on case studies in managmenet.
>
>*DM - Which in my mind says that there are hundreds, if not thousands of
>concepts and axioms in this intersection of law and management (two fairly
>diverse domains that in your example need to interoperate). You keep
>asking people to pledge allegiance to the PAR, but this, and almost
>everything you say about SUMO, indicates to me that you don't personally
>subscribe to the language in the PAR that talks about the SUO as a
>structuring mechanism for a number of domain ontologies.
I think you misunderstand me then. I agree completely with the PAR, having
written or collaborated on writing the early versions of it, so it would be
odd if I didn't. Any domain specific ontology with hundreds or thousands
of concepts, if they were not quite general, would indeed belong in a
domain-specific extension to the SUO. The domain ontology would not be
part of the SUO, but, just like the domain specific ontology work that is
going on in the OMG, that domain specific ontology might be a separate
standard of its own, which also conforms to the SUO.
>e) the ontology could support an information retrieval application that
>allows retrieval of contractual and legal documents about the structure of
>an organization. Precision would be improved beyond statistical algorithms
>by allowing the user conducting a search to specify which types of
>organizational relationships he or she is concerned with.
>
>*DM - Once again, you say "*the* ontology", which is the main point of my
>original note. I really think it is important for all of us to be thinking
>in terms of "ontolog*ies*". I now realize this is at the heart of why I
>abstained vis a vis SUMO, with a comment about methodology. Based on a
>clearer understanding now of my discomfort, I would almost surely vote no
>on the SUMO - not so much as a set of concepts and axioms that might have
>a lot of interest, but as a proposed singularity where what is needed is a
>mechanism to manage a multiplicity.
I'm sorry you feel that way. I thought the PAR was clear that we were
creating one ontology (certainly that was what was intended in the original
language) but other folks have managed to have a different
interpretation. This does seem like such a fundamental disagreement though
that we might considered splitting this group into two, one of which is
aiming for one ontology and one which is aiming for a group of
ontologies. The problem is, that since we have only one group, legitimate
proposals that have behind them one of these two different goals, are being
rejected (by individuals) just because the proposal doesn't address a goal
it wasn't intended to.
> >I'm not asserting that these proposed additions to SUMO do not address the
> >various goals of the PAR. I am just starting to think through for myself
> >how this should all work, and particularly in the context of the IFF
> >architecture that has now come under the control of the WG.
>
>sounds good.
>
>*DM - Well, maybe ... :-)
>
>Adam
>
>*DM - snipped the original note
Adam Pease
Teknowledge
(650) 424-0500 x571