Re: SUO: Re: SUO Ballot with 2 Questions
Pierre,
Matthew West summarized the reasons for a registry
very succinctly, and I quoted them in an earlier note.
Now, he has made another good point about them, which
I again quote:
MW> Why do you think it is contentious to manage the content
> of the lattice of theories? (which is all that using a registry
> means in practice?)
> Please also note that SUMO and OpenCyc are not the only potential
> contributions.
> If a single ontology is the "right" answer, putting the constituent
> ontologies into a lattice and asking questions about the differences
> is probably the best way to bring that about. Alternatively it may
> also bring about understanding as to why there should be more than
> one.
I can't detect any significant difference between what you are
asking for and what Matthew is saying that a registry can assist.
PG> But we need to know what actually is inside a system, not how they are
> marketed. We will not need to parse each systems into modules and expect that a
> structured hierarchy will emerge from clear water. The modular structure of Cyc
> was not motivated for accomodating a variety of UOs. Cyc has one upper-level
> ontology, not ten. With respect to the content, the microtheories we shall find
> there are not necessarily relevant to SUO.
The Cyc developers adopted, compared, and contrasted every ontology,
terminology, classification, textbook, and reference work about
ontology that they could get their hands on. They analyzed them,
compared them, and took what they could from them.
Today, there are many more resources available than when the Cyc
project began in 1984 or when its upper levels were reorganized
in the mid 1990s. Bateman's list not only includes the dozen or
more upper ontologies, it also includes many, many more lower
level collections. The total number of concepts that have been
cataloged and described number in the multiple millions.
Cyc, SUMO, and the other upper level ontologies are distillations
of what certain people who looked at the raw data happened to
think were important. They're smart people, and their opinions
are worth considering. But they're not infallible, and many
other smart people feel that their proposals are inadequate.
We won't know what is relevant or irrelevant unless we analyze
the content modules, relate them to one another, and ask pertinent
questions about them. That is, I believe, what all of us want to do.
> You seem again to agree that we need to know what we are looking for. In the
> mean time you don't seem to be able to leave on the side your love for the
> lattice of theory (who knows that there would be a lattice anyway in the case
> of SUO?) Building a lattice of theories is a decent goal when you want to build
> a lattice of theories. It is nearly absurd to start with this intend when all
> you want is to factor out material which will partake in the construction of a
> SUO. That we will actually need a modular structure is a gratuitous claim, and
> I think in the first place an implementation issue, irrelevant imho.
There are several very important questions we must ask about the
relationships among any ontologies and their components:
1. Are they individually consistent?
2. Are they mutually consistent?
3. Is one a subset or superset of another?
4. Are they subsets of a common superset?
If so, which one(s)?
5. Are they supersets of a common subset?
If so, which one(s)?
The answers to these questions define a generalization hierarchy
of the theories. A refinement of that hierarchy that fills in
the gaps is a lattice. Motion #2 merely assumes a hierarchy,
not a full lattice.
And a registry is nothing more nor less than a method of
indexing the modules and adding metadata about who, what,
when, where, and why any changes, updates, revisions, or
modifications were made.
> 2) registry is a baneful and not only a waste of time, it puts the issue of
> content in the background, but content is the only thing which should drive the
> SUO effort.
The content is what's in the registry. The primary addition is
the metadata that explains the content and how it is related to
other content (including references to the published literature
from which any of the axioms may have been derived).
> 3) registry based on SUMO and Opencyc is dangerous for both of them taken
> together are not representative.
I agree. That is why we also have to look at other content.
A registry is merely a convenient place to put the content.
And the metadata in the registry is a record of the analysis
and observations about that content and its interrelationships.
> 4) yes, the issue is re-use. We need to know what kind of things we want and
> come up with it. If some of the things we want are in the SUMO or OpenCyc, take
> them from here. Otherwise take them from somewhere else, otherwise, do it
> yourself.
Excellent point. That is why we need to record all the
sources available, which ones were considered, and why.
That information is part of the metadata attached to each
module in the registry.
> I am agaisnt a SUO which would be ad hoc in its content because we had to:
> 1) make room for SUMO
> 2) make room for OpenCyc
So am I. The statement that we begin with those two is simply
an acknowledgment that their developers are active members
of this group and are among the people who will be doing
the initial work. But there is no commitment to accept
any or all of the material in either one.
And if we do a good job, we mihgt be able to bring back
some of the people who were early members of the SUO back
in 2000 and 2001. Many of them, who are actively developing
other ontologies on Bateman's list, might be persuaded to
rejoin the SUO project if it had a structure in which they
could contribute their efforts. The registry is an invitation
to them to bring their work into a common framework with
the other contributions.
> and ad hoc in its structure because we had to:
> 3) use IFF
> 4) make your infinite lattice leitmotiv part of the catechism
The IFF developers are very intelligent people who are also
active members of the SUO. It makes sense to include them.
But motion #2 does not make any commitment to use all or
even anything of IFF beyond the generalization hierarchy
(which is what you get from the 5 questions above). Motion #2
does not mention a lattice or say anything about infinity.
> I hope this gives a better idea of where is the problem.
Again, I fail to see any problem here. What you and Matthew
are asking for is completely compatible.
John