Re: SUO: Ballot with 2 Questions
Tim, Pierre, and Eric,
First a comment from a supporter:
Tim King wrote:
> I am not sure that I understand this to be technically impossible...
True. I should have qualified it by saying "impossible within the
limits of available resources." As Cyc has demonstrated, 19 years
of elapsed time and 650 person-years of work has not produced an
upper ontology that everyone can accept.
> but I can imagine that a "all-or-nothing" proposal gives us no easy
> basis on which to sort out the acceptable content from the
> troublesome. Otherwise, I am with you all the way.
Yes. That is the point. We need to break down the task
into manageable steps.
Pierre Grenon wrote:
> it is not obvious how feasable or efficient this strategy would be.
> It seems to me a strategy as worthy as building a city of all possible
> houses in order to build a standard house. In this case what you
> suggest is rather to build a neighborhood or a street, with the
> provision that we could extend to a city.
A neighborhood full of houses (i.e., the couple dozen ontologies
in Bateman's list) already exists. Motion #2 merely says that
we start with the houses that are already there, register them,
and register the inventory of parts and subparts they contain.
Some of the work needed for that inventory has already been
done (i.e., the microtheories and modules of OpenCyc and SUMO).
The next step is to look at the existing houses to determine
whether any of them can contribute useful parts or substructures
for building our dream house. Unlike physical houses, we can
take pieces from an ontology without destroying the original.
If we are lucky, the existing houses have all the parts
we need, and we can build our dream house just by putting
together the right parts. More likely, we will have to go
out and buy or build some new parts. But the original
neighborhood can give us a good starting set.
> Anyway, if a standard can emerge from combinations of modules,
> how come another standard can't emerge from work on the raw material?
In principle, it could. That would be like starting with the
neighborhood of houses without having an inventory of the parts
that each house could contribute to the dream house and without
any plans or measurements that show the sizes of the parts and
how they fit together. You could do it, but it would take
a lot more time and effort.
Eric Peterson wrote:
> I'm quite encouraged to hear that our long range goals are more in line
> that I thought they were.
To be clear, I am not against the idea of having a
standard SUO if a very, very good one could be defined.
But I am very much against the idea of picking one
just for the sake of having a standard. And so far,
nobody has demonstrated anything that could even be
called good -- much less very good or very very good.
> [ELP] If your first point really is true, then the "organization" that
> the registry lattice will provide will be rather uninteresting and flat.
That is the first step, which merely lines up the neighborhood
of houses. The next step is to take the inventory of parts.
> The only two cases where I can see us getting more than one lattice
> layer (not counting top and bottom) are when:
>
> i) There is a preexisting lattice of cooperating dependant
> (micro)theories
> ii) Someone creates a monotonically altered variant of an existing
> (micro)theory
That is probably true if we are limited to just those microtheories
that the original developers chose to identify. I seriously doubt
that any of the microtheories of OpenCyc correspond to any of
SUMO's 11 modules.
To use the analogy of the neighborhood, it is unlikely that
you could take any structure the size of an entire room from
any house and copy it over to your dream house.
But you can certainly reuse lots of parts the size of bricks
and boards, and even larger parts, such as staircases, doors,
window frames, chandeliers, plumbing fixtures, and furnace.
And to extend the analogy, you could probably reuse even larger
parts from the neighborhood if you could persuade the original
designers to remodel their houses to use common substructures.
In effect, that is what motion #2 asks the SUMO and OpenCyc
developers to do: work together to determine whether they
can increase the commonality by finding and extracting common
substructures or by making adjustments to their designs that
would maximize the commonality.
The IFF developers said that they would be willing to work
with both the OpenCyc and SUO developers to take inventory
and to see what cutting and fitting has to be done.
John