Some longish comments below. (01)
Leo
_____________________________________________
Dr. Leo Obrst The MITRE Corporation, Information Semantics
lobrst@xxxxxxxxx Center for Innovative Computing & Informatics
Voice: 703-983-6770 7515 Colshire Drive, M/S H305
Fax: 703-983-1379 McLean, VA 22102-7508, USA (02)
-----Original Message-----
From: uos-convene-bounces@xxxxxxxxxxxxxxxx
[mailto:uos-convene-bounces@xxxxxxxxxxxxxxxx] On Behalf Of Bill
Andersen
Sent: Sunday, February 26, 2006 12:53 AM
To: Upper Ontology Summit convention
Subject: Re: [uos-convene] Other Approaches Too. (03)
Jim raises some good (and some IMHO mistaken) points. Seems as good
a time as any to put my $0.02 in. See below for comments. (04)
On Feb 25, 2006, at 19:08 , Schoening, James R C-E LCMC CIO/G6 wrote: (05)
> All,
>
> I agree with Steve Ray's statement in the draft communique,
>
> "We all agree the use of some formally defined common upper
> ontology is essential for semantic interoperability." (06)
I would not go so far as to say "essential". EII and EAI are not
*complete* failures today. They do work, but require a lot of
effort. The promise of (upper) ontology is that it makes the cost of
doing these things acceptable where it can't be argued that it is,
except for situations where the ROI warrants the cost. (07)
LEO: I think EII and EAI are largely successful because they take a
global view of the enterprise and attempt to address it with models,
perhaps for the first time in mainstream computing/information
technology. (TOVE and the Enterprise ontology of Uschold et al focused
on the enterprise many years ago, but mainstream computing was not
ready; unfortunately, MC moves at glacial speed and adopts new
technology very slowly if at all.) (08)
LEO: Also, I think that EAI is still largely syntactic, using tools
such as Enterprise Architect and languages such as UML for modeling,
but these are primarily human-interpreted, and aren't
machine-interpretable. Good for making pictures that perhaps we can
agree on, like Powerpoint with fine widgets. Some of us have again
proposed enterprise architecture ontologies, i.e.,
machine-interpretable theories about enterprise architectures and
enterprises to make the semantics explicit and take advantage of
automated reasoning. (09)
> I support exploring this new approach, but let's not dismiss other
> possible approaches, such as:
>
> 1. Major Leader Approach:
> a. A large player selects one CUO (after seeking input and
> consensus within an open forum)
> b. Its business partners use it and it spreads gobally
>
> 2. Consortium Leader Approach:
>
> a. A consortium of key players develop or select a CUO.
> b. It spreads globally
>
> 3. Market Momentum Approach
> a. Many players use different CUOs and the market eventually
moves
> toward one of them.
>
> 4. Consensus Approach
> Open forum seeks to develop or select a CUO.
>
> My assessment of each:
>
> All approaches need more commercial success of basic system
> ontologies. (010)
This is a key point. While I don't have a good handle on precisely
how much OWL, for example, is being used, the uses I have seen of it
outside academic contexts have, without exception, been trivial -
rarely beyond what could have been done with RDFS. As I have said in
other fora, I would like to know what about these uses constitutes
"ontology" where data models in use since the 70s and especially
since UML do not. In short, we're a long long (long) way from
talking about ubiquitous interoperability or integration. (011)
LEO: I think there is quite a bit of OWL development going on. I see
many examples of it in the intelligence community, for example.
Admittedly in the commercial world, perhaps not as much. However, OWL
being primarily a description logic, and description logics focusing
primarily on subsumption/classification reasoning, i.e., where in the
subsumptive taxonomy should your specific assertion be placed, you
really need more than OWL if you want to do interesting automated
reasoning, e.g., deductive reasoning. Then you need the OWL extension
SWRL, and SWRL is very close syntactically to Horn rules in logic
programming. OWL as a DL, as all DLs at least for me, is not enough. I
want deductive reasoning/proof. (012)
LEO: One distinction between entity-relation (or extended
entity-relation) and UML on the one hand and RDF and OWL on the other
is that the latter are primarily logics and have a defined formal
semantics. I know that UML is evolving towards a logic, but it is not
yet there, and it still has a very limited formal semantics. And to me,
being defined as a logic means that a modelling language is thereby
machine-interpretable (not machine-understandable, mind you, since
there is no understanding, and not just machine-readable and
machine-processable, since XML and UML are those), i.e., can make
inferences comparable to those a deducing human would make. (013)
> #4 IEEE SUO WG (which I chaired) tried this but achieved little
> consensus, due in part to lack of utilization of the candidate
> upper ontologies, lack of pragmatic vendor participation, (014)
(guilty?) :-D (015)
> lack of market momentum toward any one candidate, and maybe just
> because there is no one correct upper ontology. (016)
Again, I don't think there are many applications out there (at least
as far as I know) outside academia that are doing anything serious or
large enough to test the "upper-level" hypothesis. The only
commercial systems I know of that employ an upper ontology in a
central role are Cycorp's and ours (Ontology Works). We have many
projects where our upper level is being exercised and our empirical
data on development and lifecycle costs indicates that ULO makes a
big difference. As for integration we can't say definitively because
no customer has attempted to use our system for that. I can't speak
for Cycorp but I suspect they could tell similar stories. (017)
LEO: Adam Pease has told me that many developers are using SUMO. And I
know from experience that folks are using at least portions of Upper
Cyc, as indeed did we at VerticalNet, though not without difficulties
(difficulties mostly caused by rapid development of multiple domain
ontologies and the increasing need to create spanning middle ontologies
to link those domain ontologies). (018)
LEO: Also, Barry Smith talks about the vast heterogeneity of
bio/medical ontologies and their absolute need for a rationalizing and
reconciling upper ontology, which that community is slowing "getting". (019)
> #3 I don't see this happening any time soon. Stand-alone
> ontologies don't need and aren't using upper ontologies. (020)
I don't deny they aren't using them. Again, the "what's the
difference between your ontology and a data model" question arises.
I do deny that they don't need them. As I mentioned above, our
experience indicates otherwise and I believe there are good non-
empirical arguments that ULO, properly done, beats "roll-your-own"
ontology every time where the roles of ontologies extend beyond those
played today by data models. (021)
LEO: Data models primarily focus on "local" semantics, not "real world"
semantics, i.e., the very narrow view from a need to organize data for
specific purposes and applications in a given part of an organization.
The usual database modelling paradigm is to 1) create a conceptual data
model (schema) using ER or extended ER and lately UML, which by the way
is the closest the database community gets to explicit semantics, 2)
refine that to a logical data model (schema), 3) refine the latter to a
physical data model (schema). The reduction from ER to the purely
relational physical model throws away all the semantics orginally
captured in the conceptual model, which maximizes efficiency of
access/querying and generality of joinability, but trashes semantics.
The only semantics you have left is in the picture of the conceptual
model and the English documentation in the data dictionary. Humans have
to look at the pictures and read the text. There is no
machine-interpretability. Also, of course, much of the semantics of
databases is not in the schema but in the code of surrounding
application programs, where it must be interpreted by human programmers
and also is not extractable nor reusable except by cut-and-paste
methods. There millions of database schemas and object models out
there: how many are similar? How are they comparable? How many are
reusable? (022)
Jim makes good points. All hinges on whether ULO is useful. I
believe it is but we are a long way from providing anything more than
anecdotal evidence. As that evidence comes in it will become
apparent exactly what role ULO is playing and exactly what benefits
it brings. (023)
LEO: I don't see any solution for semantic interoperability/integration
that excludes the use of an upper ontology (or a lattice of theories,
if you will, some of which are upper theories). You can indeed create
domain ontologies ex nihilo and for stand-alone use. We have all
created these because you have to start somewhere. However, my
experience is that even in developing a domain ontology, you spend a
good portion of your time inventing middle and upper ontology
constructs (time, location, process, behavior, state, change,
mereology, etc.) that you absolutely need for your domain ontology. And
you end up creating these quickly, with very little thought beyond how
your current domain ontology needs that construct, and having very
little specialized knowledge (time pts or invervals, 3D vs. 4D,
mereology/mereotopology/constitution, etc.) about better ways of
creating those constructs. So you build those notions in quickly and
badly. (024)
LEO: It's only when you've either done the latter ex nihilo,
stand-alone ontologizing for a while and have observed the waste and
inefficiency that drives you to increased reuse, or need to integrate
two or more such ontologies because at a certain point human resources
needs to interact with logistics, or planning needs to interact with
purchasing, or buying needs to interact with selling, that you
understand the value of integrative ontologies, middle ontologies, and
upper ontologies. Who needs semantic interoperability if you blissfully
live in your own domain or application, and don't need to interact with
others? The need increases for explicit semantic models as you move
from very local to systems of systems to enterprise to community levels
of interaction (and I've got a slide showing that, which I attach [1]).
And the more explicit semantics you need, the more those loosely
coupled islands of semantics need to be embedded in integrative,
middle, and upper ontologies. (025)
LEO: [1] Obrst, Leo. 2006. Tightness of Coupling and Semantic
Explicitness, Ontolog briefing, slide 5, Jan. 12, 2006,
http://ontolog.cim3.net/file/resource/presentation/LeoObrst_20060112/On
tologySpectrumSemanticModels--LeoObrst_20060112.ppt. Addiitonally, the
attached is animated, annotated. (026)
Finally, let me suggest this as a topic that the Summit might spend
some time on. I would be very happy to participate in such a
discussion. (027)
.bill (028)
Bill Andersen (andersen@xxxxxxxxxxxxxxxxx)
Chief Scientist
Ontology Works, Inc. (www.ontologyworks.com)
3600 O'Donnell Street, Suite 600
Baltimore, MD 21224
Office: 410-675-1201
Cell: 443-858-6444 (029)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/uos-convene/
To Post: mailto:uos-convene@xxxxxxxxxxxxxxxx
Community Portal: http://ontolog.cim3.net/
Shared Files:
http://ontolog.cim3.net/file/work/UpperOntologySummit/uos-convene/
Community Wiki:
http://ontolog.cim3.net/cgi-bin/wiki.pl?UpperOntologySummit (030)
Tightness of Coupling & Semantic Explicitness.ppt
Description: Tightness of Coupling & Semantic Explicitness.ppt
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/uos-convene/
To Post: mailto:uos-convene@xxxxxxxxxxxxxxxx
Community Portal: http://ontolog.cim3.net/
Shared Files: http://ontolog.cim3.net/file/work/UpperOntologySummit/uos-convene/
Community Wiki: http://ontolog.cim3.net/cgi-bin/wiki.pl?UpperOntologySummit (01)
|