Re: Types, sets, and relations

• Mail folder: Interlingua Mail
• Next message: phayes@cs.uiuc.edu: "Re: Practical effects of all this discussion"
• Previous message: sowa: "Re: Practical effects of all this discussion"
• In-reply-to: phayes@cs.uiuc.edu: "Re: Types, sets, and relations"

Message-id: <199304282241.AA08927@quark.isi.edu>
Date: Wed, 28 Apr 1993 15:40:31 -0800
To: phayes@cs.uiuc.edu
From: macgregor@ISI.EDU
Subject: Re: Types, sets, and relations
Cc: interlingua@ISI.EDU

Pat,

> For example,
>suppose I were to simply ignore it, and just translate all types into
>monadic predicates, without telling you. How would you ever discover I had
>done this? What difference would it make?
>(If the answer is just that I wouldn't be able to translate back
>satisfactorily, assume I can do that. I might have a 2nd order predicate
>called 'McGregorThinksItsAType'.)

You have (almost) answered your own question. All I need to get the
functionality I want is a second-order type. However,
"MacGregorThinksItsAType" is not a good name. First of all, I don't
think that I could convince anyone else to include a name like that in
a standard :-).

My reason for wanting a predicate with a standard name (for example, we
might call it "type" or "sort") is not for translating back from KIF to
the original input language -- its for translating to some other
language whose type facility is reasonably compatible with (in my case)
Loom's. So, for example, if I wanted to use KIF as a vehicle for
translating a Loom KB to a CG KB, I might decide that the two notions
of "Type" in this pair of systems are sufficiently compatible that I
could safely use that predicate as a means for distinguishing between
types and properties (or whatever each of the two systems chooses to
call these internally).

Your first message, detailing multiple distinct semantics for the
notion of a "sort", provides good evidence for NOT assigning a
semantics to "sort" (or "type") in KIF. But, it also provides an
argument for including a sort or type predicate in KIF -- the
proliferation of sorted systems would seem to argue for the relevance
of "sort" as a conceptual building block, independent of a specific
semantics.

Perhaps a better vehicle for introducing conceptual KR building blocks
into KIF would be defining a limited but official "frame" ontology (in
the Gruber sense) as an auxiliary part of KIF. I would like to see
things like Type, Role, Proposition, Belief, etc. included. These are
examples of terms which everyone uses, but which don't have a
universally-accepted semantics. My conjecture is that the use of these
terms by various systems is close enough that from a practical
standpoint, there is benefit to be gained by, for example, mapping from
the type system in one KR system to the type system in another KR
system, or from one belief system to another.

My second conjecture is that there will never be universal agreement on
the meaning of these terms. Hence, waiting until we reach agreement is
not a good decision procedure for determining when to add these kinds
of terms to KIF.

- Bob

Robert M. MacGregor                                     macgregor@isi.edu
USC/ISI, 4676 Admiralty Way, Marina del Rey, CA 90292      (310) 822-1511