SUO: Re: Consensus?
Eric,
I wouldn't say that the joint motion proposes
a "new" ontology. Instead, I would rather call
it a library of modules, which may be derived
from any source, with OpenCyc and SUMO as two
of the principal sources.
There are several advantages to having such
a library:
1. Flexibility: Any developer has the option
to select as many or as few modules as
necessary for a particular project without
making a commitment to adopt the entire
"mind set" of the full SUMO or Cyc world
view. This option allows developers to
have world views (say, on the nature of
time, space, causality, human emotions,
cognitive models, etc.) that are very
different from the ones that are built
into SUMO or Cyc, while still taking
advantage of predefined modules for
noncontroversial topics.
2. Legacy software: A complete ontology,
such as Cyc or SUMO makes many commitments
that are unlikely to be identical to those
of existing software. In order to use those
systems in conjunction with new ontology-based
software, it is essential to represent the
implicit ontology of the existing systems.
That requirement makes the flexibility of
point #1 essential.
3. Testing and reliability: "Unit testing" of
small modules is a prerequiste for and usually
much simpler than testing a large system.
The same is true of ontologies, and most of
the smaller modules of the library are likely
to be much more reliable and much better
tested than a large package such as Cyc or SUMO.
4. Consistency: This is actually part of point #3,
but I'm breaking it out because it is especially
important. The basic issue is that neither Cyc
nor SUMO has ever been proved to be consistent.
But many, if not most of the "microtheories" or
modules of SUMO and Cyc can be or have been proved
to be consistent. For systems that are expected
to run 24x7x52 hours per year without a hiccup,
time-limited "consistency checks" are not an
option. Such systems could not use any module
or package that had not be proved consistent.
I must also emphasize that there are *no* disadvantages
to the library of modules, since all of SUMO and all
of OpenCyc can also be present in the library as single
modules. People who need or want the full package can
have it, but they also have the option of choosing smaller
parts if they prefer.
Some comments on your comments:
>[ELP] I don't at all in principle oppose the "new" ontology - as long as
>we are willing to wait for the extra merging time and the time for the
>proving of the new ontology. I have a hard time envisioning such a
>newly crafted/merged ontology to be up to the level of robustness of
>OpenCyc in less than a couple or three or five years of hard work - and
>that is with strong participation.
>
The first point to make is that many modules are
not "new" since both SUMO and OpenCyc were built
from contributions of smaller theories from
various sources. Therefore, many of the modules
in the library can simply be lifted out of the
existing packages.
There may be new "crafted/merged" modules that are
constructed from combinations of SUMO and OpenCyc,
and one might have to wait for them. But a major
advantage of a library is that it is modular:
new modules can be added one at a time whenever
they become available.
Furthermore, a larger system can *never* be more
robust or consistent than its components. The modules
(or microtheories) that make up SUMO or OpenCyc are
guaranteed to be at least as robust or consistent
as the whole (and probably much more so).
>This may argue for just picking A or B rather than constructing a C.
>
Someone who is starting from scratch on a new
project with no previous commitments may be well
advised to pick a complete package. But see
the arguments above for good reasons why many
developers couldn't, wouldn't, or shouldn't do so.
>[ELP] To me, the key entities in a SUO are the domain entities such as
>person, place, event, etc. These items I consider to be first class
>data rather than data about data. The term "meta" is diluting and
>becoming more meaningless with every passing day. Meta to me entails
>axioms about axioms, attribute facets like in Ontolingua, or some other
>such notational variant of the same notion.
>
I agree with you on this point.
John