COMMENT ⊗   VALID 00002 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00002 00002	%napole[f89,jmc]		Napoleon and Wellington hearing of each others deaths
C00007 ENDMK
C⊗;
%napole[f89,jmc]		Napoleon and Wellington hearing of each others deaths

Napoleon hearing of Wellington's death:

1)	∀p e(time hears-of(p,e) > time e

2)	∀p e (performer hears-of(p,e) = p)

3)	∀e (time e < time death performer e)

From these we compute

time death Wellington < time hears-of(Napoleon,death Wellington)

		      < time death performer hears-of(Napoleon,death Wellington)

		      < time death Napoleon,

and this contradicts the historical fact

time death Wellington > time death Napoleon.

(In this discussion we take for granted facts and inferences about the ordering
of times.

	Now we weaken the axioms by introducing  occurs e, asserting
that the event  e  actually occurs.  We weaken 3) getting

3')	∀e(occurs e ⊃ time e < time death performer e).

Now we can infer

	¬occurs hears-of(Napoleon,death Wellington)

by deducing a contradiction from the premiss

	occurs hears-of(Napoleon, death Wellington).

	Further weakening of the axioms is called for in order to get
axioms general enough to put in a common sense database, e.g. CYC.
However, we would like to preserve the compactness of the above proof.
It uses only simplification and propositional calculus once the
proposition  occurs hears-of(Napoleon,death Wellington)  has been
formed and  time death Napoleon < time death Wellington  has
been retrieved.  Introspection suggests that any acceptable set
of axioms should admit an equally straightforward reasoning process.
There will be more premisses in the axioms, but their satisfaction
should be a matter of simplification and database retrieval.

Weakenings:

1) Not all events have performers.

2) hears-of(Napoleon,death Wellington)

cannot be a rich object, i.e. it doesn't specify whether Napoleon
was sitting or standing.
hears-of(Wellington,death Napoleon) turns out to be a rich object,
since it occurs.  However, before the system knows which occurs
it ought to be able to treat the two similarly.


Wellington hearing of the death of Napoleon:

This seems simpler.  Without considering the special features of
the relation between Wellington and Napoleon, we could infer that
Wellington would have heard the big news.  Moreover, that's the
relation most usefully formalized.  Nevertheless, it should also
be possible to gild the lily and give additional reasons.

	bignews e ∧ active p ⊃ hears-of(p,e)

holds in contexts with definite time.  Also

	vip Napoleon

	vip p ⊃ bignews death p.

	In contexts without definite time we need something like


bignews e ∧ time(e, active  p) ⊃ time (e,hears-of(p,e))


Lenat and Guha concerns:

Doug expects looking at structure of abnormalities to pay off.  He
would replace  occurs  by an abnormality predicate.

Guha worries that if an event or other rich object is described
in a way that incorporates some wrong detail, then the non-existence
of the object will be inferred and all the correct inferences lost.