Dear John, (01)
I'm pleased to see we seem to be converging. A few further points below. (02)
> Please note that I never said that my observations constitute a proof
> that no upper-level ontology could ever be useful. I have always said
> that it is a worthwhile *research* project. (03)
[MW] I'm glad to hear you say that explicitly, because the impression I
think most people will have had from what you have said on the subject is
that upper ontologies are a waste of time.
>
> I have also said that the largest existing ontologies (Cyc with about
> 600,000 concept types and the Japanese EDR with about 400,000) have
> cost multi-millions of dollars to develop, and they have not be able
> to support a sufficient number of applications to recover even a
> fraction of that investment. But other companies have implemented
> successful applications based on smaller domain ontologies.
>
> These observations do not prove that a good upper level is impossible
> to discover, but they do suggest three observations:
>
> 1. It's not easy to discover a good upper ontology and to integrate
> it with a large number of special domain ontologies. (04)
[MW] I disagree. Given that the main (valid) differences between upper
ontologies arise from the ontological commitments made. The number of
ontological commitments where you need to make choices is relatively small,
and many of the combinations are unattractive, leaving say half a dozen to a
dozen that are reasonably coherent and defensible. Integrating domain
ontologies with your chosen upper ontology is largely a matter of
translating it into the paradigm that those commitments comprise. Tedious
perhaps, but not difficult.
>
> 2. Even with a large ontology, it's not easy to develop a methodology
> for using it effectively. (05)
[MW] That depends on what you a trying to do with it. The only killer app I
know of is systems integration, and I have even co-edited an ISO standard
methodology (ISO 18876 - Integration of Industrial Data for Exchange Access
and Sharing) for doing that, so there is no need to develop a methodology,
just use what exists.
>
> 3. However, the successful applications of smaller ontologies
> indicate
> that further research on integrating and relating them would be
> useful. (06)
[MW] I don't think this is a matter of research. It is one of finding the
business case and getting on with it.
>
> Therefore, I believe that we should do three kinds of things:
>
> 1. Continue the research on all levels of the ontologies and on
> methodologies for integrating them, relating them, and using them. (07)
[MW] Yes of course.
>
> 2. Use the available smaller ontologies to implement successful and
> profitable applications. (08)
[MW] Of course. No point worrying about bread tomorrow when you don't have
bread today.
>
> 3. Plow back some of the profits from #2 to support #1. (09)
[MW] I would rather make part of 2 contributing to 1 - then the customer
pays, as they should.
>
> Some comments:
>
> [MW] If a mapping is not precise, then it has very little value in my
> experience.
>
> I agree. And that is my primary argument against a large development
> project based on primitives. I'm not against research on primitives,
> but all the currently proposed primitives are very poorly defined. (010)
[MW] I'm inclined to agree.
>
> [MW] However, there will be some upper ontology that is implicit in
> what they have done whether they like it or not. So I think I need
> to be clear and spell out what I think the essence of an upper ontology
> is. I think the following are the key elements of an upper ontology...
>
> I agree. But again, discovering the commonalities implicit in many
> smaller ontologies is a research project. (011)
[MW] No it isn't. It is part of the integration project that needs the
result. (012)
> My proposal of a lattice
> (or at least a partial ordering) of theories would provide a framework
> for showing and recording all the implicit relationships. (013)
[MW] I agree. I am a strong supporter of your lattice of theories ideas.
>
> [MW] I would welcome either an explanation of how you can avoid using
> these relations [listed in the previous note] and making commitments
> in these areas...
>
> I certainly agree that all those relations (and others) are important,
> and I would emphasize them, not avoid them. (014)
[MW] Good. So we do need upper ontologies then, even as part of domain
ontologies.
>
> Some analysis would be needed to discover all implicit relations, but
> the partial ordering would allow any commonalities to be promoted up
> the hierarchy whenever they are discovered. (015)
[MW] Yes I agree layering what I think of as different levels of genericity
and construction is important.
>
> Furthermore, as new ontologies are designed by combinations of parts
> of the older ones, the method of combining them would automatically
> show the generalization, specialization, and sibling relations among
> the ontologies and parts of ontologies. (016)
[MW] Yes, but one of the challenges is that for any new ontology you need
what Whitehead calls a "speculative metaphysics" which is essentially the
ontological commitments and the core categories and relations that provides
the framework for a coherent picture of the world. Any time you pick a
different framework you have to produce a mapping to any other frameworks
that have been chosen by others. This is expensive.
>
> [MW] I suggest that these will necessarily include the elements I
> set out above, so they do include an upper ontology - you just don't
> want to call it that.
>
> I certainly do want to call it an upper level. But I believe that
> the upper levels should be *discovered* through analysis, not by
> any a priori imposition of somebody's pet theory. (017)
[MW] Here we disagree. I agree that you should avoid unnecessary
commitments, but in the areas I set out there are choices you have to make,
if what they cover is in scope. You can "discover" the different
possibilities and the results of making the different choices, but for
example 3D or 4D does not appear by "discovery" in the traditional sense of
emerging. They are models we impose on the world, and it is necessary both
to recognise that is what we do (because we might be wrong) and not shrink
from (because it is necessary).
>
> The advantage of the partial ordering is that it is infinitely
> extensible. It can accommodate any and all current, proposed, or
> future theories at any levels -- upper, lower, or middle. (018)
[MW] Yes, I think this is a good way to capture the different possible
combinations of ontological commitments and their consequences.
>
> But it does not force any a priori assumptions about which
> theories are better or worse than any others. (019)
[MW] I agree that, at this stage at least, one should not be trying to pick
winners. (020)
> The users are
> free to choose whichever they find useful, and the patterns
> of usage can suggest which directions to emphasize. The best
> de jure standards are the ones that recognize and canonize
> the most successful de facto standards. (021)
[MW] Sometimes you need to create a standard so that there can be products
based on it, and without a standard there are no products (e.g. CD players).
>
> [MW] ISO 15926 has definitely paid out many times over. It is
> not as large as Cyc, but there is nothing to stop it becoming larger.
> The real difference is that it was developed with the solving of
> particular problems in mind, but so that it could be extended
> to solve other problems.
>
> That is a good way to develop ontologies. ISO 15926 would
> certainly be put in the hierarchy (lattice or partial ordering). (022)
[MW] Yes, that is what I would look for.
> (023)
Regards (024)
Matthew West
Information Junction
Tel: +44 560 302 3685
Mobile: +44 750 3385279
matthew.west@xxxxxxxxxxxxxxxxxxxxxxxxx
http://www.informationjunction.co.uk/
http://www.matthew-west.org.uk/ (025)
This email originates from Information Junction Ltd. Registered in England
and Wales No. 6632177.
Registered office: 2 Brookside, Meadow Way, Letchworth Garden City,
Hertfordshire, SG6 3JE. (026)
> [MW in note to Pat C] Frankly until you can show that there are
> uses that give significant benefits, you have not actually done
> anything useful (though it may be interesting)...
>
> That is precisely my objection to Pat's proposal. As I said,
> it's an interesting low-budget research proposal, but not yet
> suitable as a high-budget ($30 million) development proposal.
>
> John
>
>
> _________________________________________________________________
> 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
> (027)
_________________________________________________________________
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 (028)
|