Thread Links Date Links
Thread Prev Thread Next Thread Index Date Prev Date Next Date Index

Re: SUO: Re: Proposed SUO Content Outline




"Frederick N. Chase" <fnc@mitre.org>:
>Thank you, Jon and Chris, for working with my posting.
>
>Below I'm expanding a little on what I said, in light of your comments.
>(And no doubt further exposing the unevenness in
>my understanding of our subject.)
>
>
>I was trying to further affirm the notion that,
>distasteful and complexifying though it may be,
>it will be essential for us to provide a place for
>theories, causal bubbles, axiomatizations, alternative courses,
>views, domains of interest (DOIs), viewpoints, or whatever they be
>termed
>which offer some sort of isolation one from another.
>(I recognize that some of these may be fundamentally different from
>others.)

Right, I agree this is needed. The current preoccupation with 
'namespaces' in Web applications is a related concern.

>The kinds of DOIs I had in mind include the following examples,
>though I'm not sure all all of these are suitable for an
>IEEE SUO, or for a chunk that can be merged into an IEEE SUO.

Heres another, much more mundane, example. Many temporal databases 
assume that time is discrete, and all durations are multiples of some 
atomic time-tick (often a millisecond). Many temporal ontologies used 
for physical reasoning assume, on the contrary, that time is 
continuous, or at least dense. These assumptions are in direct 
logical contradiction. However, the contradiction is usually harmless 
in practice because no sane reasoner would use the density axiom on 
an atomic time-interval; and it is easy to repair the inconsistency 
by putting a protective condition on the density axiom if needed.

More seriously (and more subtly), many temporal ontologies assume 
that time is the real line, but many physical ontologies (including 
the SUO) use some version of the Allen interval algebra. These 
assumptions are also in direct conflict. They can be reconciled, but 
only at some conceptual cost; for example, by assuming that an 
interval is not identical to the set of points it contains, or by 
assuming that all intervals are half-open. This kind of repair work 
involves a much more serious re-thinking of the overall temporal 
ontology.

<snip>

>I was rejecting the "Into the Sky"/"Into the Ground" dead-end
>that Jon Awbrey and Chris Menzel describe.

This 'dead-end' is largely an illusion of Jon's, apparently due to 
his not knowing anything about the state of the art in reasoning 
systems. The point being that a reasoner does not (in fact, cannot 
possibly) immediately infer all the valid consequences of its 
knowledge, and the same machinery that enables it to detect validity 
of an inference also detects inconsistencies in its own Kbase. So 
they can be (and often are) written to be self-checking. Cyc is 
constantly checking itself for internal contradictions, and often 
finds them, without going utterly mad. (Jon is also quite wrong about 
nonmonotonic reasoning, by the way, which is not a complete 'into the 
sky' abandonment of consistency, but an organised way of allowing 
consistency to accumulate as a result of a series of corrections to 
reasonable assumptions. It needs to be treated with special care to 
avoid a kind of unstable logical oscillation, but the dynamics of 
this are now well understood, and there are elegant and coherent 
semantic theories which account for the logical properties of such 
logics.)

>Rejecting it in favor of some sort of hack involving an
>isolation-of-consequences or scoping-of-names notion and
>implicitly asking that anyone who has some idea of how to do this step
>in and
>give the idea additional legitimacy (if it deserves it).

There is an entire literature now on techniques for debugging 
ontologies by rooting out latent contradictions. Can we just take 
this for granted and move on?

Pat Hayes

---------------------------------------------------------------------
IHMC					(850)434 8903   home
40 South Alcaniz St.			(850)202 4416   office
Pensacola,  FL 32501			(850)202 4440   fax
phayes@ai.uwf.edu 
http://www.coginst.uwf.edu/~phayes