SUO: RE: (ELP's summary of MRW's standards experience) Was: A NEW FUNDAMENTALLY DIFFERENT FORMAL MOTION...
Dear Eric,
See responses below.
Matthew West
Principal Consultant
Shell Information Technology International Limited
Shell Centre, London SE1 7NA, United Kingdom
Tel: +44 20 7934 4490 Other Tel: +44 7796 336538
Email: matthew.west@shell.com
Internet: http://www.shell.com
> -----Original Message-----
> From: Eric Peterson [mailto:epeterson@CCAAVA.com]
> Sent: 18 June 2003 18:20
> To: West, Matthew R SITI-ITPSIE; John F. Sowa; Mike Pool;
> apease@ks.teknowledge.com; clegg@cyc.com; John DeOliveira
> Cc: Patrick Cassidy; standard-upper-ontology@ieee.org
> Subject: RE: (ELP's summary of MRW's standards experience) Was: A NEW
> FUNDAMENTALLY DIFFERENT FORMAL MOTION...
>
>
> Thanks for your reply.
>
> Please see below:
>
> <snip>
>
> > > Although your standards group was large, only a hand full of
> > > its members
> > > claimed to have design experience. The other members were domain
> > > experts.
> >
> > MW: No, I was quoting the background and foreground participants
> > with "design" (my word would be analysis) experience. There was/is a
> much
> > larger diaspora of domain experts, say hundreds. We ended up working
> > in increasing circles by a combination of peer review and agreement.
> > >
> > > I think most of us in the SUO would claim ontology design
> experience.
> >
> > MW: Yes. Some 10 years ago in EPISTLE we had face to face
> meetings of
> > 30-40 people that felt very like what is happening here. It took a
> > while to build trust and for the first 6 and then 3 to
> emerge who were
> > trusted to lead the work technically.
>
> [ELP] Perhaps we should agree to differ on their being any
> hope of that
> in this group.
MW: Perhaps. As I see it we are in the early stages of this. And as you
know I talk in terms of years rather than months.
>
> > >
> > > I, therefore, see much more potential for design disagreement
> > > in a group
> > > like the SUO. And since we haven't shown a great amount of unity,
> we
> > > can't yet claim that having more designers is better.
> >
> > MW: Not as long as everyone wants to be in charge. One of
> the reasons
> I
> > support John Sowa's proposal is that I think it will support the
> > development
> > of convergence, but allow a large number of people to stay on board
> whilst
> > the consensus emerges.
>
> [ELP] This group does not yet agree on whether we are merging
> ontologies
> into a single standard ontology. But my first motion has
> been seconded
> by John S., so maybe we can vote and stop glorifying the
> notion of what
> I will call a hydra standard. Hopefully the hydra that is
> just helping
> us to merge is innocuous or maybe helpful.
MW: I would see merging things that are accidentally different as
important. I also see recognising real differences as important too.
>
> I wonder how many of us actually read the extensive web site
> on what the
> repository for which you all voted?
>
MW: I know I struggle. I am still trying to get to grips with Category
Theory and its strange (unnecessarily so in my view) language. However,
I have grasped enough to see its potential, and I am happy to follow
Robert's lead on this (I didn't create the language used for the EPISTLE
Core Model either). (So perhaps leaders and followers in certain areas
are starting to emerge).
>
> > > ______________________________________________________________
> > > __________
> > > ___
> > >
> > > Complexity-wise, I see your small design group as having conducted
> an
> > > admirably large design effort. But technology-wise it was a well
> > > understood technical exercise (relational DB design with
> a bit of OO
> > > mixed in).
> >
> > MW: There was no relational DB design, we developed an entity
> relationship
> > model.
>
> [ELP] I've long observed that, in terms of practical usage in the
> industry, the two are equivalent. So I'm sorry for mischaracterizing
> your work. But I claim that perhaps the most fundamental and
> important
> part of an OO or ontological design is its relational subset.
MW: In fact EXPRESS has its own implementation environment, but in EPISTLE
we have as much looked for a relational implementation.
MW: Having a recognised way of implementing in a database is an
advantage. However, the database design would have (probably) between
1 and 10 tables, rather than the 201 entity types. So you can see that
the database design process is someting separate from determining the
ontology.
> In fact,
> the inheritance relation is typically demoted to just another relation
> in non-frame-based ontological systems. So be proud of the relational
> part of your work. Stand up and be counted ;^)
MW: I am indeed proud of the work (of others) in doing some extremely
smart implementation work. I just separate that from ontology design.
>
> And since I have a hard time envisioning most of your users using a
> non-relational model, I rather assume that one of the "multiple
> versions" of your standard of which you have spoken, is a relational
> model extracted from your OO model. That sounds like more or
> less what
> I've been trying to get my customer thinking about.
MW: We don't have a standard database implementation. It is a point for
competitive advantage or different approaches for different purposes.
The rule is that the physical model should support the conceptual data
model.
>
> This is roughly equivalent to a DL based ontology. However, the
> > tough bit was recognising the need for, finding and agreeing to a
> paradigm
> > (four dimensionalism in our case) that meant that we could
> develop the
> > model in a regular and consistent way so that where there was
> disagreement
> > we had more than opinion to fall back on. The problem I see
> here is an
> > unwillingness to do the hard work of getting down to the
> bedrock so we
> > can build solid foundations.
>
> The acoustic returns that I'm getting indicate the bedrock is a couple
> of miles down below marine clay. That's why I'm suggesting starting
> with an existing ontology.
MW: Its really not far beneath our feet, and some is exposed. The bits
I've found are First Order Logic, Set Theory (ies), Category Theory,
Mereology, and you need to add a theory of individuals. Mine is
4 dimensionalism. It says we exist in a space-time manifold, we have
temporal as well as spatial parts, and if two things have the same
spatiotemporal extent, they are identical.
MW: 3 dimensionalism gets more complicated, but you actually need to
define "your" 3 dimensionalism in a similar way to my definition of
4 dimensionalism, otherwise it is open to different interpretations.
>
> > We all know what happened to the house built
> > on sand.
>
> [ELP] You want far more that the simple initial foundation
> that I want.
> You want to load it up with axiom. You want n^2 mappings between
> existing ontologies. You don't like the existing candidates, so you
> seem to want to jump out into untested waters. You are truly brave.
MW: I just want to start with the foundations, and build it up
incrementally.
>
> 28 For which of you, intending to build a tower, sitteth not
> down first,
> and counteth the cost, whether he have sufficient to finish it?
>
> 29 Lest haply, after he hath laid the foundation, and is not able to
> finish it, all that behold it begin to mock him,
>
> 30 Saying, This man began to build, and was not able to finish.
>
MW: I have noticed in many mediteranian countries that many families
build their own home. They start with the foundations and first floor,
and you will see reinforcing rods comming out of the flat roof. Later
(when they can afford it or perhaps the family expands) they add
another floor, and so on.
>
> Isn't this group composed chiefly of folk spending discretionary work
> time and/or personal time?
MW: Yes.
>
> This group has nothing resembling a work breakdown/plan that I could
> analyze and potentially agree with. I think we share this concern.
MW: My plan is to wait for Robert to do his stuff and try to keep the
group behind what he is doing. It is also to try to get him to present
his stuff in sufficiently ordinary langauge that mere mortals will
know how to use it, and can poke and prod at it. That will do for the next
year or so.
MW: In the mean time we should discuss the foundations for a 3D ontology
so we can have a principled and grounded base for that. I think Nicola's
paper(s) and Ted Sider's book are good places to start.
MW: We will then be able to review the OpenCYC and SUMO (at least) work
against that foundation, and practice comparative ontology by building
up a 3D and 4D ontology together (there will be different insights
probably from each).
MW: We may find we need more than these two, and there possibly will be
different theories of particular areas within these. I think that is OK
and should be judged on utility (different theories for different
purposes, but justified and catalogued).
>
> But my SWAGS come up orders of magnitude short of us having the
> resources to do what you want to do in a reasonable time.
MW: Reasonable against what standard? A particular project you have in
mind? The you are right. Pick SUMO the EPISTLE Core Model, DOLCE, OpenCYC
or something similar and do the best you can.
> But since you
> have the hope and the long term view, I nominate you to come up with
> such a work break down. No matter how far off it turns out to be, it
> should prove to be a much needed straw-person to start that
> dialogue and
> pushing us toward deciding what our charter means in
> practical terms and
> whether we believe in it.
MW: See above for an outline.
>
> > >
> > > SUO is attempting to design something that has:
> > > - Inference rules
> > > - Free-form FOL axioms
> > > - A design intended to properly anticipate all lower level designs
> > > depending on it.
> > > - Little in the way of canonical technical approaches in these
> areas.
> >
> > MW: We were only missing the free form FOL axioms.
>
> [ELP] It sounds like your rules had the full power of Horn logic.
MW: Not sure.
>
> > >
> > > I hope we agree that SUO is doing something much more complex and
> much
> > > less proven.
> >
> > MW: No, I just see it as taking the next incremental step from where
> we
> > were. As far as a 4D ontology is concerned we have a decent
> foundation
> > and framework. All it needs is some additional work to add a next
> layer
> > of axioms. Once we get past working out the best representation, we
> need
> > do little more than examine the axioms floating around in this arena
> > and reverse engineer them into a 4D ontology. Not trivial, but not
> > outrageous either.
>
> [ELP] Do they bottle this optimism somewhere? I'd love a
> shot of it;^)
MW: You haven't seen how powerful 4D analysis can be, and how good some
of the people here are. But I am only looking to add what is not
controversial, and establish a process which may never end.
>
> Actually, I admire much of your patience and long-term view. And I
> shared a differently organized brand of it. I just think that there
> needs to be some line of demarcation out there between standards work
> and development. If we are going to call it all standards
> work, can we
> agree to invent two new words that map to what I call
> standards work and
> what I call development?
MW: Tell me what you have in mind as the distinction.
>
> Where do you draw the line between standards work and development.
MW: Development is doing something specific - developing a product, or
supporting a project. Standards work is developing stuff that can be
reused later. Often by picking up experience from projects and making it
more general, sometimes by noting a problem and doing research to get
to a good solution. We are in the latter category here.
>
> >
> > Now for those wishing to develop a 3D ontology, you are mostly
> > back where we were 10 years ago with no clear foundation to
> build on.
> > I see some hope in the work Nicola is doing with DOLCE, and some
> similar
> > efforts to start building an ontology on a principled
> basis. But until
> > those ideas take hold, you are doomed to go round in circles.
> > > ______________________________________________________________
> > > __________
> > > ___
> > > Size-wise you list ~200 entity/relations.
> >
> > MW: Plus some 50,000 concepts as reference data ...
>
> [ELP] Sorry, I assumed you were talking about instances/tuples so I
> didn't count them as being part of the data model.
MW: They are instances/tuples, but the data model includes a meta-model
so there is a merging between data and data model, and the reference
data is essentially an extension - in fact much of the data model is/
will be also held as reference data.
>
> > >
> > > SUO could easily end up with 3000. That's not even counting
> > > non-structural (non-DL) axioms.
> >
> > MW: Which is why it is so important to get solid foundations.
>
> [ELP] Right. Can you be more concrete about what you don't like about
> Opencyc and SUMO - in the context of them being solid foundations?
MW: They do not have equivalent statements to the 4D ones I gave above.
Ted Sider's book and Nicola's paper talk about some of the issues where
you need clarity about your position. Unclarity leads to ambiguity
which leads to different interpretations being possible, both in extending
the ontology and holes in inferencing over it.
>
> > > ______________________________________________________________
> > > __________
> > > ___
> > > ______________________________________________________________
> > > __________
> > > ___
> > >
> > > I must admit to inaccurately extrapolating from my own standards
> > > experience. When John Sowa agreed with my "no standard
> without a de
> > > facto standard" mantra and cited a lot of personal standards
> > > experience,
> > > I thought I was making a safe assumption in believing it to be an
> > > uncontroversial SUO view
> > > (http://grouper.ieee.org/groups/suo/email/msg09209.html).
> But that
> no
> > > longer appears to me to be so.
> > >
> > > But, having said that, the moral I take from your standards effort
> is
> > > that a standards group has a chance at crafting an acceptable
> standard
> > > if (i) Much/most of the design work is already done
> >
> > MW: Don't see where you get this from. We had a couple of
> start points
> > not unlike OpenCyc and SUMO. One was clearly better, so we started
> from
> > there. It was principalled, (better than SUMO and OpenCyc) but not
> well
> > enough. along the way we threw out everything we started
> with - mostly
> > when we finally adopted 4D.
>
> [ELP] I was hoping you would quantify what was already done,
MW: Our start point had a heritage of about 20 man years effort, from
an internal (Shell) Corporate Data Modelling effort, but
it was not planned to be in a straight line, it just worked out that
way.
> but if you
> completely threw out any trace of the previous work (including the
> experienced gained) - never mind.
MW: No of course not. I only mean that over a 10 year period, by the
time we finished you could not see much if anything of the start point.
>
> >
> > > (ii) the designers
> > > are unified via technological belief or by hierarchy,
> >
> > MW: We certainly had technical differences. But we developed an
> approach
> > to
> > resolve them.
>
> [ELP] Do tell.
MW: it is really being disciplined about managing an issues log.
During ballot cycles/ review periods issues against the standard can be
raised. These must be logged. You then work diligently to resolve those
issues, seeking agreemtn of those who raised them that they have been
resolved, or noting that this is too difficult at the present time.
This visibility inspires confidence, and means people feel engaged
rather than ignored. operated properly it is a very good way to
stimulate an improvement culture.
>
> >
> > > (iii) the
> > > designers are funded,
> >
> > MW: Correct. The funding level is not enormous though.
>
> [ELP] So you three weren't close to full time?
MW: The technical work has probably been about 25 days a year for each
of us over the last 5 years or so, perhaps more before that (this is for
the data model only, there was another team working on the reference
data).
>
> >
> > > (iv) the remaining design work is not
> > > technologically controversial,
> >
> > MW: Well no-one has done anything like we have done in EPISTLE, and
> > so nearly everything we have done has been considered controversial.
> > As one example, no-one before has combined data model with reference
> > data.
>
> [ELP] You've got me curious. What are your customers doing with your
> standardized rules?
MW: Most of the usage is as a taxonomy, so that people can exchange data
across system and organisational boundaries. The rules are really only
around having a coherent taxonomy. At one point we had constraints in the
model that meant that large groups of classes had to be duplicated, so
we had to work out what we were doing wrong. What is unusual about our work
is the wide scope with which the data model can cope. Most data models are
restricted to relatively specific applications, and I am not aware of an
Enterprise Data Model for a large organisation (Fortune 500) even.
>
> >
> > > (v) the design work is not
> > > technologically complex,
> >
> > MW: Don't know where you get this from.
>
> [ELP] Sounds like quite a system.
MW: Its a data model not a system. There are some challenges in implementing
it because the data that you would perhaps expect to find in adjacent columns
might be spread across 4-5 entity types.
>
> >
> > > (vi) and that the domain is not too large in
> > > entity/relation count.
> >
> > MW: 50,000 concepts and growing is small eh?
>
> [ELP] So these are domain subclasses that you are keeping
> separate from
> an upper ontology? Can you give me a URL to a particular specimen?
MW: The top 10,000 are publicly available at:
http://edmserver.epmtech.jotne.com/rdl/rdl_core_public.html
Some caveats. It is mostly a taxonomy. The focus of content is the
process/oil industry (they are paying). Also it is currently against a
previous version of the data model, with which there were some problems
for the RDL, so some stuff is not where it should be. But otherwise
take a look.
The aim is to move it into a formal ISO Register over the next couple
of years and set up a process for incremental addition/improvement.
having different people able to propose content is important to make
significant progress, but equally means that it has to be clear what
the principles are so that different people doing the same thing would
produce the same result.
>
> >
> > > But this is not to claim that your
> > > model was not
> > > a complex model within the relative simplicity of the
> > > relational model.
> >
> > MW: Entity Relationship is NOT relational.
>
> [ELP] I would avoid using the "ER" term without qualifying it
> as to the
> beyond-relational features you use. Otherwise, boatloads of us will
> assume the typical relational usage.
MW: Subtype/Supertype relationships is our main excursion, but a relational
design is also all about looking at access paths and performance and such
like, which we pay no heed to. Somebody elses problem.
>
> >
> > > Your model sounds rather impressive.
> > >
> > > Moving forward, I don't think we are going to agree on the
> > > general need
> > > for a prior de fact standard. But I'm hoping that we can agree
> trying
> > > to simplify the SUO process.
> >
> > MW: We need to find a METHODOLOGY that allows us to move forward in
> > something approaching a straight line - if slowly.
>
> [ELP] Amen.
>
> <snip>
> [ELP]
> Best,
>
> -Eric
>