Patrick communicated to ONTAC-WG
General Discussion, a note starting with
1) Hub? I think that ultimately
one COSMO will best serve the purposes of interoperability, and for me the
question is how to get there -- to start with one upper ontology and elaborate
it by mapping to the multiple KCSs used by ONTACWG members, or to start with
with several competent ontologies, and to merge them. My suggestion was to
try both approaches simultaneously, i.e. in one part to investigate the
potential for merger by formalizing the UMLS semantic network using each
candidate upper ontology separately, and to compare the resulting
fragments of each upper ontology to determine how closely those
parts are related and just how easy a merger would be. In a second
part, the FEA-RMO and DoD Upper taxonomy could also be formalized with respect
to one of the upper ontologies, chosen by some criteria such as those
suggested by Eric Peterson. That is, we would not prejudge the issue
of whether or not to choose one existing hub from the start, but to gather
more information to determine whether a merger would be so difficult that
choosing one existing ontology is the only practical course. The question
of using an existing ontology as a hub for formalization of includes the
question of what criteria would be used to choose the hub, and any thoughts on
that issue would be welcome right now.
Respectfully, he seems stuck on
wanting to do something that seems right but which has principled arguments against his persistent
suggestions. His voice is similar
to others in the standards processes, but this voice fails to appreciate not only viewpoints but also the social reality as expressed by
several other statement today to the ONTAC-WG General Discussion. Several months ago, Patrick and I
had phone conversations about this and I represented some of the conversation at
http://www.ontologystream.com/beads/nationalDebate/188.htm
where I conjecture about
the error in wanting to have ontology that supports general types of inference
of the type observed only in human reasoning and not observed from AI.
At
http://www.ontologystream.com/beads/nationalDebate/191.htm
I further
expressed my view of his position and
the error that is made in persisting with certain expectations.
I made the mistake of
mentioning where he worked (at a large government think-tank-consulting entity)
and the discussion became about that (my mistake.) I have edited this to remove this
mention. I do not think that he is
fully aware of the grounding that stratified ontology has in the natural
sciences, and that he is focused on a specific viewpoint regarding what makes
business and consulting activities work in the current IT consulting world.
The persistence of his
arguments does not seem to be persuadable.
As such, I feel that there cannot be a discussion of alternatives to the
notion of a single large general-purpose ontological model.
****
I am struck with the notion that
the proposition of an alignment of two Protege type ontologies, such
as FEA-RMO and (say a OpenCyc upper ontology ? if there is one actually
available to the public), is a proposition that ignores the reality of
interpretation and viewpoint.
Accepting something like the Dublin Core as
an available ontology is a different type
of path towards interoperability. Dublin core is accepted by many because
it is simple and specifies some really
useful information standards, mostly related to the act of publishing
information.
But basic concepts to be included in his common large ontology
are related to "the reality" of
things like duration and composition (the new attack was experienced from Monday
through Thursday). To discuss the detailed nature of "this new attack" within a
large community would require something beyond the non-functional properties
proposed as a web services ontology model. So suppose the attack has novel
elements or elements that are known but function in an entirely new (and
surprising way).
The OASIS
work on OASIS Reference Model for Service Oriented
Architectures
as discussed
briefly at
http://www.ontologystream.com/beads/nationalDebate/201.htm
creates a
higher level abstraction (framework) in which web services can be made
interoperable (either as a hard wired relationship between interacting systems
or as a just in time aggregation activity).
Human communication, at its best,
iteratively spreads the knowledge of "a new attack" within a large community...
but in the case of 9-11 and the American public the eventual knowledge came to
reflect different individual and community
viewpoints.
The alternative that is proposed by some, is
a stratified ontology where some "observed" set of invariance (semantic primes)
are commonly used by individuals and other processes (including pattern
recognition algorithms) to express (in real time) emergent
ontological models.
Like in physical chemistry, the diversity of
possible chemical compounds is sufficient to cause the chemistry at a moment
(the ontology at that moment); and to adapt to the response degeneracy (Gerald
Edelman's term for many to many mappings) seen as living systems deal with
structure-function choices.
I am struck by why this notion of stratified
ontology is not ever discussed within any of the standards processes?
Stratified ontology can be grounded in physical science (chemistry), in speech
production by humans (phonemes), and in language use and generation (Tom Adi's
work)
http://www.bcngroup.org/beadgames/generativeMethodology/AdiStructuredOntology-PartI.htm
Adi's work has been around and in software
form for two decades, but remains largely outside the Academy and outside of the
consulting/standards processes.
But one can look at the Zackman Framework
and Sowa's semantic primitives to see the stratified approach, even if not
recognized as such.
This
absence of discussion about stratified theory is why I am also bothered by
insistence on a large common "fixed" and controling ontology.
Paul
Prueitt
703-981-2676
Gary
Berg-Cross has articulated a number of
legitimate issues concerning the method for moving toward a Common Semantic
Model. His points are diverse, and think it best to respond
to some of these points by describing my own views -- which may not be
those ultimately chosen by the COSMO-WG as a
whole.
(1) Hub? I think that ultimately
one COSMO will best serve the purposes of interoperability, and for me the
question is how to get there -- to start with one upper ontology and elaborate
it by mapping to the multiple KCSs used by ONTACWG members, or to start with
with several competent ontologies, and to merge them. My suggestion was
to try both approaches simultaneously, i.e. in one part to investigate the
potential for merger by formalizing the UMLS semantic network using each
candidate upper ontology separately, and to compare the resulting
fragments of each upper ontology to determine how closely those
parts are related and just how easy a merger would be. In a second
part, the FEA-RMO and DoD Upper taxonomy could also be formalized with respect
to one of the upper ontologies, chosen by some criteria such as those
suggested by Eric Peterson. That is, we would not prejudge the
issue of whether or not to choose one existing hub from the start, but to
gather more information to determine whether a merger would be so difficult
that choosing one existing ontology is the only practical course. The
question of using an existing ontology as a hub for formalization of includes
the question of what criteria would be used to choose the hub, and any
thoughts on that issue would be welcome right
now.
Concerning Gary's feeling
that:
>> I
don?t see how this could be done with merging some existing ontologies and
various people have talked about using UMLS DOLCE/BFO, SUMO, OpenCyc, ISO 15926, FEA-RMO and the DoD Core
Taxonomy.
It is important to distinguish the formal upper
ontologies from the FEA-RMO and DoD Core taxonomy. The latter two
are candidates for formalization with respect to one of the upper ontologies,
they would not serve as upper ontologies on the same level as the
others.
It is not clear a priori without careful
investigation that any one of the potential hubs has everything that is in all
of the others, and does not need supplementation. The formalization of
the UMLS-SN is a non-trivial but tractable test case that should provide some
objective information to help evaluate the candidates and estimate the
difficulty of merger. I am hoping that we can begin discussing specifics
soon.
(2) The FEA-RMO and DoD core taxonomy do indeed have
problems if looked at as starting ontologies. The DoD Core doesn't
pretend to be an ontology, and the FEA-RMO is, as Gary points out, closer to a
Reference Model, though expressed in ontology format. There are
nevertheless good reasons to try to connect these to the UMLS. (a) The
FEA-RMO is an active project, which may well form part of any common ontology
ultimately developed for use within that federal government. There are
expectations that the SICoP will help in some way in the development of
that model. The ONTACWG could be helpful in making recommendations on
the best way to formalize the FEA-RMO and align it with other knowledge
representation efforts within the government, even if much of the work will in
fact be done by contractors. I think it would be unfortunate if we
ignored it, since we can bring a broad perspective that can make ontologies
developed from or related to the FEA-RMO more functional. (b) The
DoD core taxonomy is still an active project, and its purpose is very much
similar to that of the COSMO, to create a hub with which domain taxonomies in
DoD can be aligned and related. But it is not intended to function as an
ontology for logical inferencing. By defining the terms of the DoD
Core using concepts from the COSMO, we can make it more useful for logical
inferencing, and the mapping of the DoD Core will have a multiplicative
effect because of its already existing relations with other
taxonomies. If there are hierarchical links in the DoD core that
appear invalid form an inheritance point of view, those links would not appear
in the formalized version, but the individual terms could still be
defined by reference to the COSMO. So the lack of formality of the DoD
Core would make the precise formalization more time-consuming, but would not
affect the formalized version, which would have only valid
relations Combined with the UMLS-SN, these two knowledge
classifications provide a diverse set of topics to serve as test cases for
building a COSMO and defining domain knowledge classifications by means of the
COSMO. To be feasible for a volunteer group yet broad enough
to provide a realistic test of the potential for semantic integration, I think
that this set of knowledge classifications contains the proper balance of
complexity and restricted subject matter, and they also relate directly to the
purpose of the SICoP, the parent of the ONTACWG.
(3) The ANSI X3/T2 Ad hoc group on ontology
standards:
I participated in the activity of this
group until it faded away around 1998 or so. As I recall, there
were some productive discussions of issues, but at least one proposal to
obtain funding for the activity was rejected. In the absence of support,
it was not possible for participants to spend enough time to resolve the very
complex technical issues, and participants returned to work on the projects
for which they could get funding. Neither OpenCyc nor SUMO nor
DOLCE were available at that time.
Another smaller project was
funded by the Klaus Tschira foundation, to bring together 20 or so ontologists
for two weeks in 1998 to come to agreement on the structure of the topmost
levels of an upper ontology. Most (not all) participants seemed to think
that useful progress was made, but that a lot more additional work was
needed. I do not think that a formal report was prepared with details of
the agreements reached. A later effort,
the IEEE-SUO project (2001 - present), did get some financial support, and the
SUMO ontology was created from that. The SUMO had some input from the
IEEE group, but was mostly built in-house by Teknowledge. The absence of
funding for anyone outside of Teknowledge left in place the same problems
that bedeviled the ANSI group, namely that the concerns of multiple
communities could not be addressed effectively, and the resulting
ontology represented the views predominantly of a small group, who of
necessity had to move forward quickly to provide the deliverables, regardless
of whether external participants agreed with the results. This
problem was not primarily a result of any lack of interest of the Teknowledge
builders for input from the wider community, but was a structural problem
resulting from the amount and distribution of funding. This project
illustrated one major barrier to agreement -- that in the absence of
applications that could test alternative suggested ontological
representations, and without a mechanism for efficiently including alternative
representations, an ontology, even built to some extent collaboratively,
will inevitably have a lot of elements that are unsatisfactory to one or
another potential user, or lack some they want.
The resources that were discussed by
the ANSI group then are still valid, but have been supplemented by additional
more developed ontologies and lexical resources. We have invited
participation by all groups, including those that created those older
resources, but as a volunteer group, it is only possible for us to make
substantive efforts using the resources that are of interest to ONTACWG
participants. The WordNet could be investigated directly if there were
any members with a special interest in that classification. But it will
be included in any case indirectly, because there are mappings of SUMO and Cyc
to the WordNet.
There never has been adequate funding
for a substantial collaborative project involving multiple groups, to find
agreement on a common upper ontology, even though proposals have been made
since the time of the ANSI efforts. We can do what is possible without
funding, and hope to encourage agencies to provide funding at some point
soon. The cBio project may help indirectly, if any people participating
in that project include some work with upper ontologies -- or better yet, with
the COSMO -- in their proposals.
The ONTACWG, without funding, can make
progress only to a limited extent, but we have more experience and more
examples now to work with than the earlier projects did. So our task
will be easier than the earlier ones. A plausible goal would be, not to
create a common upper ontology de novo, but to investigate within some limited
compass how to choose or adopt one or more existing upper ontologies to
serve as the hub to which multiple domain knowledge classifications can
be related. As mentioned above, I think the ONTACWG can serve a valuable
function by investigating the alternative approaches of choosing one upper
ontology versus merging several. Even within the limited scope I
suggested, this will take a lot of detailed work. To address the problem
of accommodating the views of different users, my suggestion has been to
structure the COSMO so that it will incorporate all views considered useful by
participants. The result is likely to be a lattice of theories.
The precise nature of the lattice that will be needed is not agreed on at
present. Discovering the details of what is needed to accommodate
multiple ontological views will be, I think, one of the important results of
the attempt to formalize the UMLS-SN versus multiple upper
ontologies.
(4) DOLCE and OntoClean
The DOLCE ontology is one of the
upper ontologies that the COSMO-WG will be investigating as a base for the
formalization of the UMLS-SN, and perhaps of the other KCSs that we are
interested in. DOLCE and BFO are the base ontologies that Olivier
Bodenreider and Lowell Vizenor are planning to use in their already ongoing
effort to formalize the UMLS-SN. The other upper ontologies of interest
to ONTACWG participants will be investigated if any of the participants decide
to take the lead in that effort. So the OntoClean method will be
considered as part of the COSMO-WG agenda. There are other formal
methodologies that have been discussed in the literature, such as Michael
Gruninger's "Semantic integration through invariants" in the Spring 2005 AI
magazine. I hope that it will also will be possible to use such
ideas in the COSMO effort.
(5) Other
approaches
The suggestions made thus far do
not preclude alternatives. It is, I think, potentially useful for
multiple approaches to be tried at the same time, because we are in a position
to share intermediate results on a regular basis and take advantage of each
other's experience as soon as anything useful is discovered or proposed, and
posted to our Wiki. The ONTACWG is not a competition, but a
collaboration. So the inefficiencies of trying alternative approaches in
traditional research, where one learns of the results of other work a
year or so after a project has been completed, should not occur in the work we
are discussing. If any participants have a particular interest in trying
a specific approach toward relating KCSs to each other, the ONTACWG
environment should provide feedback and support that will make individual
efforts more productive.
The paper by Doerr, Hunter, and Lagoze starts off expressing a sentiment
that is also shared by myself and a lot of others, that some common top-level
ontology will be needed for accurate semantic interoperability. And some
of the specific observations they make as a result of their study are, I
believe, valid. But as has been pointed out, their proposed merged
ontology does have some problems, and falls far short of the best developed
ontologies that are already available. A one-on-one alignment of that
kind may be useful for applications that are restricted to two or a
small group of communities, but our interest is much broader. I
think we can do a lot better.
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 at mitre.org
I wanted to follow
up Eric Ps earlier message about a hub approach to building our common
ontology. I think that his
questions and issues got side tracked by the responses to Roy?s "general
ontology".
Eric was curious
about ?how pervasive those anti-hub feelings really are. ? I?m not of one feeling on this issue,
since I think it is complex and would welcome some discussion of some of
it. Eric had particular ideas
on Dr. Sowa?s sub-sumption lattice idea, but I haven?t heard responses to
that and perhaps others can respond to it.
For myself, I
could imagine going a hub or modular approach depending on the quality of
the hub. I?d have to be
convinced that it was doable with our resources, and would want to know the
?seed? for it and the process or development. I don?t see how this could be
done with merging some existing ontologies and various people have talked
about using UMLS DOLCE/BFO, SUMO, OpenCyc, ISO 15926, FEA-RMO and the DoD Core
Taxonomy. Assuming even this as
a start I have some issues and a approach to discuss as a strawman.
The FEA-RMO is an Ontology of a Reference Model and not of an
actual domain such as health.
It seems quite hard to connect to this to others.
Also, the DoD
taxonomy, in my opinion, has the degree of problems that Barry and John
pointed out in the "general ontology" so it may not be easy to
assimilate. We might start
without trying to merge these in and also might start with the best 2 or 3
as candidates to seed an effort.
Another
point or question concerns leveraging the experience of past efforts. Back seven years or more there was
an effort by the
ANSI Ad Hoc group to construct a standard, called the Reference
Ontology. They had a five-step
approach for the following
- Upper levels (approx. 100,000 terms):
Bring into correspondence (to align) the terms of a small number of
selected large-scale ontologies (eventual size approx. 100,000 items). Do
so inclusively; that is, create a result in which users can choose which
of the component ontologies' terms they wish to see and use.
- Domain models (under 2,000 terms each):
Link into this Ontology selected domain-specific ontologies, developed to
support reasoning about time, space, physics, geography, etc. Do so
inclusively; allow the linkage of various different models of time, space,
etc.
- Access tools: Create easy-to-use tools for
Ontology access and extension.
- Dissemination: Place the resulting
Reference Ontology on the Web, freely available.
- Theoretical
basis: In
ongoing work, have a team of highly qualified individuals comb through the
Ontology to find powerful generalizations, to weed out unnecessary and
inconsistent items, and to create a maximal factoring of the upper levels
of the Ontology.
Seems
quite similar to what we are talking about. Whatever happened? Did it fail because it didn?t have
an upper ontology?
They listed the following are candidate sources
for terms to be included into their ?merged Reference Ontology? and a few of
these (UMLS , CyC) have been mentioned as a base for us too :
- USC/ISI: Pangloss Ontology
SENSUS approx. 70,000 terms, general coverage, little detail,
taxonomization supports Natural Language applications.
- Princeton: WordNet approx.
70,000 terms, general coverage, little detail, taxonomized on Naive
Semantics / Cognitive Science principles.
- CYCorp: Upper portion of
CYC ontology approx. 2,500 terms, general coverage, little detail,
taxonomized on Naive Semantics / AI principles. Later additions may
include more of the 40,000-odd terms currently in CYC.
- EDR: Upper portion of EDR
concept ontology approx. 1,000 terms, general coverage, medium detail,
taxonomized for Natural Language applications. Later additions may include
more of the approx. 400,000 terms in the EDR concept lexicon.
- New
Mexico State University: MIKROKOSMOS approx.
4,000 terms, general coverage, detailed, taxonomized for Natural Language
applications.
- European Union:
EuroWordNetóunder construction; probably approx. 50,000 terms,
little detail, taxonomized on Naive Semantics / Cognitive Science
principles.
- LXT Inc.: UMLS medical
ontology exceeds 50,000 terms, medium detail, taxonomized for medical
reasoning applications.
Perhaps some of these should
also be on our list if they have ?matured?.
A last point/issue concerns
alignment between our starting sources and how to start on this. Martin Doerr and others did some
work reported in ?Towards a
Core Ontology for Information Integration? and described the comparison and convergence of 2
ontologies using the OntoClean
approach. (Guarino, N. and
Welty, C., ?Evaluating ontological decisions with OntoClean,? Communiations of the ACM, 45
(2), pp. 61-65, 2002,)
This uses an analyses of
top-level ontological distinctions related to:
1. instantiation versus
membership
2. part-of and mereological
axioms
3. extensionality
4. connection
5. location and extension
6. co-extension, co-connection
7. unity, singularity and
plurality
8. dependence/independence
The claim is that the OntoClean
approach ?enables: the detection of concept definitions that are lacking in clarity or rigidity;
the justification of valid sub-sumption relations; and the detection of invalid sub-sumption
declarations. ?
Would it be useful to start
looking at the matchup of some of our ?seed? ontologies in this way (or a
better way that might be proposed)? The Doerr paper discusses the process of
finding common conceptualizations by equivalency between a concept of
?Temporality? on Ont1 and ?Temporal Entity? in Ont2 or ?Action? and ?Activity? etc. Some of the foundation
concepts. It might be useful to
make our discussions concrete if we are planning on using a merger of UMLS
DOLCE/BFO, SUMO, OpenCyc, ISO 15926
etc.
Regards,
_________________________________________________________________
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 (01)
|