Re: Eliminating variables

sowa <>
Date: Sun, 20 Feb 94 17:45:27 EST
From: sowa <>
Message-id: <>
To:, interlingua@ISI.EDU
Subject: Re: Eliminating variables
Gerard Ellis made an interesting point about conceptual graphs
in comparison to some of the earlier semantic nets, which showed
every possible IS-A link as a connecting line.

My basic guideline in developing CGs was to stay as close as
possible to the natural language structure, while still maintaining
a completely unambiguous syntax that can be directly translated to
predicate calculus.   The CGs make frequent use of "type labels"
like MAN, CAT, HOUSE and proper names like John or Mary.  Those
labels behave like long-term variable names that are used and reused
so often that people learn to memorize them.  CGs also have some
use of "coreference labels", which tend to occur with about the
same frequency and in the same places as pronouns in English.

What CGs and English both eliminate, however, is the proliferation
of short-term variables for linking relations to their arguments.
I won't, however, claim this as a unique feature of CGs, because
there are other versions of semantic networks that have a roughly
similar balance of direct links and indirect labels.

John Sowa