Eric -
I believe that your point is quite relevant, that the precise
nature of how theories will map to one another is an important issue
for us, and that correlating axioms in different source ontologies will
be a necessary step to determine this relation, assuming that we are
not starting from ontologies already known to be consistent with each
other (which we aren't).
But the n^2 mapping problem should only bedevil us at an early
phase, when we are still evaluating alternative high-level ontologies.
As I understand the earlier responses regarding the direction of the
COSMO effort, most of the group is reluctant to chose a single upper
ontology from the start, and would prefer an experimental break-in
phase when alternatives are being explored. The formalization of the
UMLS semantic network provides an opportunity to do such comparisons
(DOLCE/BFO, SUMO, OpenCyc, ISO 15926) with a problem of restricted but
meaningful size. I think there is a likelihood that the lattice of
theories that emerges from an attempt to reconcile different upper
ontologies in this test case will have large areas of agreement and
smaller areas of disagreement. But that is what we need to discover,
and this group is superbly qualified to investigate such issues and to
conduct a study of this kind.
And a project starting from a DOLCE-like small upper ontology and
expanding rapidly into specialized areas can be conducted in parallel
with the UMLS formalization. For this case, we might be able to
include the FEA-RMO and a mapping to the DoD Core Taxonomy, as well as
the UMLS. Although this may involve some seemingly redundant effort,
it does in fact address two different questions, and, with the results
at each phase available on our site for evaluation and comment, we
should be able to avoid genuine inefficiencies that come from
independent projects that are unaware of each other's intermediate
results.
I also find the notion of extracting whole class trees from existing
well-constructed ontologies to be an attractive method. I think there
is a lot to like in the OpenCyc class hierarchy, for example in the
sections that would fit under the DOLCE "particulars" category -- and
the documentation is excellent. Trees under "personWithOccupation",
"Organization", Paths", "Artifact" and "Event" appear well-structured
to me, and could be incorporated as is, with supplementation (and
perhaps some reinterpretation) where ONTACWG members believe that their
own domain concepts do not match precisely the intended meanings of
seemingly similar concepts in the growing ontology. The relations in
Cyc appear to me to be worth using selectively. But there is also a
lot of valid structure in the SUMO and its extensions, as well as
axioms, many of which can be reused in the Ontology Works IODE
environment. I don't expect to find actual logical contradiction, but
there will probably be subtle differences and overlap which will
require adjustment, merger, or translation, if we will take advantage
of both.
Our problem is, of course, our very limited time to work on complex
details. I will be recommending to the owners of projects I work on to
allow much of the content to be made public for ONTACWG use, as it
could become more valuable to them by supplementation from the work of
others. I hope that we can get into substantive construction very
quickly, and pushing forward with two different methodologies should
allow contributors to work within the methodology most suitable to
their own existing projects.
Another issue is the format for the ontologies being constructed.
To take advantage of all of the expertise in the ONTACWG, it would be
useful to have versions of any ontology we build in KIF and OWL as well
as the IODE kfl language. We know that the higher-order relations,
functions, and axioms will not be usable in the OWL environment unless
supplemented with SWRL, and even then it will be less expressive than
KIF. But the non-OWL-native information can be preserved in text
strings to allow round-trip conversion. So, a final question: does any
of the ONTACWG members know of anyone who would be willing to work on
scripts translating OWL <-> KIF, KIF <-> kfl, and kfl <-> OWL? Those
among us familiar with any two of those languages can help in the
translation. (01)
Pat (02)
Patrick Cassidy
MITRE Corporation
260 Industrial Way
Eatontown, NJ 07724
Mail Stop: MNJE
Phone: 732-578-6340
Cell: 908-565-4053
Fax: 732-578-6012
Email: pcassidy@xxxxxxxxx (03)
-----Original Message-----
From: ontac-forum-bounces@xxxxxxxxxxxxxx
[mailto:ontac-forum-bounces@xxxxxxxxxxxxxx] On Behalf Of Peterson, Eric
Sent: Friday, November 11, 2005 11:17 AM
To: ONTAC-WG General Discussion
Subject: Our central COSMO focus (was RE: [ontac-forum] Surveyed
Ontology"Library" Systems -- parts) (04)
Hi Pat (and group); (05)
My focus at work happens to be more fine-grained than the partial
ordering of theories. (06)
But, perhaps there is no disagreement, because I claim that any partial
ordering of theories must either be (i) declared by fiat (like
ontolingua's "uses" relation or Cyc's "genlMt" relation; or (ii)
determined by a complete reconciliatory mapping between potentially all
axioms of the respective theories. Declarative partial ordering
presupposes that the two theories are already consistent with one
another where one theory is "using" the other and where both theories
share a common content model and meta-model. (07)
I claim that the theories that "use" one another must be "consistent",
abide by a common model and, therefore, don't need to be mapped to
another. But I claim that such a endeavor (the reconciliation of
already reconciled theories) is less relevant to this group. If that
is
true, we must determine partial ordering of theories by first
mapping/reconciling their respective axioms. (08)
Because my customer started with a DOLCE-derived artifact, our ontology
lacks the depth and breadth of some of the larger ontologies. (09)
I claim that our work must have near-term relevance. And for this
group's work to be directly relevant to my customer, we would need to
make sure that any creation of a lattice of theories is focused on the
mapping of pairs of axioms to one another - in such a way as to "grow"
a
larger effective ontology. (010)
For us, this means grafting (via mapping) subtrees of key content into
DOLCE from other ontologies. I'm happy to view this a part of a larger
n^2 everything-to-everything mapping. I say n^2 because no one voiced
agreement with my proposal to agree on mapping to a single hub ontology
starting point. (011)
By making such fine-grained mappings we will be contributing to the
partial ordering of theories by mapping individual axioms to one
another. Only after such mapping, I claim, is partial ordering of
theories embodying differing models possible. (012)
Best, (013)
-Eric (014)
> -----Original Message-----
> From: ontac-forum-bounces@xxxxxxxxxxxxxx
[mailto:ontac-forum-
> bounces@xxxxxxxxxxxxxx] On Behalf Of Cassidy, Patrick J.
> Sent: Tuesday, November 01, 2005 10:31 AM
> To: ONTAC-WG General Discussion
> Subject: RE: [ontac-forum] Surveyed Ontology "Library"
Systems -
> - parts
>
> John's outline is, I think, a good clear summary of the
general
> goals
> that I also believe this group can make progress toward. I
> would just
> add one additional point:
>
> Since interoperability of different systems will hinge
on
the
> clear
> specification of how their conceptual models are related, a
> registry
> that will serve for our purposes should have built-in
mechanisms
> for
> specifying at some level of detail how different ontologies
> relate to
> each other. At the lowest level, no analysis may have been
made
> to
> determine how one ontology relates to others in the
registry,
> and it
> will stand on its own in isolation. At another level, a
domain
> ontology may have been built explicitly to be defined by
and
to
> conform
> to the theories in some upper ontology in the registry.
There
> can be
> partial alignment as well, and some method would be helpful
to
> specify
> in a usable way how the ontologies in the registry relate
to
> each
> other; two domain ontologies, for example, may be logically
> consistent
> except with respect to some restricted uses in a particular
> context.
> Dagobert Soergel has taken the initiative to begin
organizing
> the
> ONTACWG subgroup that will study what our requirements are
for
> registries, so that we can make recommendations to the
groups
> that are
> actually building registry systems. He will be sending a
note
> to the
> list soon.
> The important distinction between the proposed Common
> Semantic Model
> and a simple registry of ontologies is that to be a
component of
> the
> COSMO, the logical relations of a theory must be specified
> clearly, in
> the manner John describes, or perhaps some other at least
as
> precise,
> to the others in the COSMO, so that the relations between
> concepts in
> different theories, and particularly identity, will be
> recognizable and
> usable for automated reasoning.
> This does not preclude inclusion in our registry of
Knowledge
> Classification Systems that have not yet been related to
others
> by such
> logical specification. Such KCSs will serve as a knowledge
> resource
> for extension of the COSMO, and if they can be related to
other
> KCSs in
> any way, may also help to improve both the breadth and
accuracy
> of the
> COSMO, but also searching capabilities for search tools
that
use
> these
> classifications.
>
> Pat
>
>
> Patrick Cassidy
> MITRE Corporation
> 260 Industrial Way
> Eatontown, NJ 07724
> Mail Stop: MNJE
> Phone: 732-578-6340
> Cell: 908-565-4053
> Fax: 732-578-6012
> Email: pcassidy@xxxxxxxxx
>
>
> -----Original Message-----
> From: ontac-forum-bounces@xxxxxxxxxxxxxx
> [mailto:ontac-forum-bounces@xxxxxxxxxxxxxx] On Behalf Of
John F.
> Sowa
> Sent: Tuesday, November 01, 2005 10:02 AM
> To: ONTAC-WG General Discussion
> Subject: Re: [ontac-forum] Surveyed Ontology "Library"
Systems -
> - parts
>
> Nicolas, Pat, Barry, et al.,
>
> This thread is getting into issues that are covered
> by the proposal for a lattice of theories:
>
> 1. All theories that anybody might propose about
> any subject whatever would be registered in a
> standard form in a metadata registry. Registering
> something does *not* imply any official status
> other than a commitment to save it in a convenient
> place for other people to examine it, use it, and
> comment on it.
>
> 2. Some theories in the registry would be more general
> and more widely reusable than others. Those are the
> ones that would eventually become the core of many,
> if not most practical ontologies. But there would be
> no need for an a priori blessing or canonization of
> any particular theory. Instead, the users would
> "vote by their feet", so to speak, in deciding for
> themselves which ones to choose for any particular
> application. The various choices and patterns of
> use and reuse would be added to the commentary in
> the registry.
>
> 3. In order to keep track of how theories are related
> to one another it is essential to show how they
> can be derived from or be converted into one another
> by the AGM operators for belief revision: contraction,
> expansion, and revision.
>
> 4. The three AGM operators define a lattice, in which
> the partial ordering defined by specialization and its
> inverse, generalization: expansion adds axioms to a
> theory to make it more specialized; contraction
deletes
> axioms from a theory to make it more generalized; and
> revision does contraction followed by expansion in
> order to move from one theory to another, which is a
> sibling of a common parent.
>
> To use the example of part-whole relations, there are large
> numbers of axioms for many different variations. See, for
> example, the excellent book by Peter Simons called _Parts_,
> which goes into great detail about many different
> axiomatizations
> and their relationships to one another. Peter did not
organize
> the theories in a lattice, but it would be possible to do
so.
>
> In summary, we could adopt the current work on metadata
> registries as a means of registering theories and making
them
> available for further use, reuse, commentary, and analysis.
> One important aspect of the analysis would be to
demonstrate
> how the various theories are related by the three AGM
> operators (to which I suggest a fourth operator called
> "analogy", which renames the predicates in a theory while
> preserving the implicational structure).
>
> The result of the analysis would be a step-by-step
> construction of a hierarchy of theories, ordered by
> specialization/generalization. An important aspect of
> the registry would be the ability to comment on the
> theories to show which ones are more widely used or
> more relevant to various kinds of applications.
>
> In short, analysis demonstrates the theoretical
relationships
> among the theories, and commentary demonstrates the
practical
> patterns of use and reuse for various applications. Both
> are necessary for a growing and evolving system.
>
> John Sowa
>
>
________________________________________________________________
> _
> 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/OntologyTaxonomyCoordinatin
> gWG
>
>
________________________________________________________________
> _
> 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 (015)
_________________________________________________________________
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/OntologyTaxonomyCoordinatin
gWG (016)
_________________________________________________________________
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 (017)
|