Re: The semantics of new-edi

Fritz Lehmann <>
Date: Fri, 23 Dec 1994 20:54:55 -0800
From: Fritz Lehmann <>
Message-id: <>
Subject: Re: The semantics of new-edi
Precedence: bulk
Dear Pat,

     The guy had specifically asked about "the technology of
conceptual graphs", not about AI in general, and I answered with
my estimation of its capabilities.  I have repeatedly, in my
series of messages to these EDI people, added that KIF, HOL,
SNePS, KLONEs, predicate logic, CYC-EL, etc. are all good candidate
languages.  I happen to like conceptual graphs at present because
they do not obscure (as variable-laden Peano-Russell notation
does) the inherent graph-theoretic nature of logic -- especially
the facts that model subsumption is homomorphism of directed
hypergraphs, and that quantifier precedence is nesting of negations.

     Also, as a technical matter, your statement:
"All CG's are, is a notation equivalent to FOL in representational power."
is incorrect.  CGs require type signatures for all relations, have
a notion of "canonicity" based on (non-truth-preseving) "joins" in
the database sense, and they have an ontological restriction on
the degree of relational expansion ("Stabbed(Brutus, Caesar)" is not
an acceptable CG, but "AGT(Stabbing1,Brutus) & VICTIM(Stabbing1,Caesar)"
is).  The latter allows "cases" to occur in a relational type hierarchy.
FOL makes no such ontological commitment (although it presumes the
distinction between individuals and predicates and usually presumes that
"something exists" -- dispensably).  CGs _use_ something like a notation
equivalent to FOL (actually weakly higher-order), but it's not right to say
they _are_ just that notation.  CGs are more restrictive than general
FOL, in the sense just mentioned.

     On another matter, I hope we can settle that event/object/process
business in a practical way.  I pretty much agree with CYC and you that
a general spatiotemporal "process" is common to, and underlies, any notions
of event versus object.  So I agree with Lenat that a person is a process.
I also agree that distinctions (whether natural language based or not)
that have no inferential consequences are not very important for us.  But
I don't think the event/object is like this, though.  Events need
participants (usually) in a way that objects don't; and objects (usually)
last longer.  Can we lock theorists in room overnight, releasing them
only when they agree?

    Your long reply to my long message on this was very garbled in transit.
I printed (but did not save) a copy.  If you have it on disk by any chance,
could you repost it with short line-lengths and without the funny
equal-signs and insertions?

                          Yours truly,   Fritz Lehmann
GRANDAI Software, 4282 Sandburg Way, Irvine, CA 92715, U.S.A.
Tel:(714)-733-0566  Fax:(714)-733-0506