Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!watmath!clyde!burl!mgnetp!ihnp4!cbosgd!mhuxl!ulysses!unc!mcnc!decvax!decwrl!amd70!fortune!hpda!hplabs!sri-unix!AMSLER@SRI-AI.ARPA
From: AMSLER@SRI-AI.ARPA
Newsgroups: net.ai
Subject: Spelling Correction vs. Fact Correction
Message-ID: <1248@sri-arpa.UUCP>
Date: Mon, 25-Jun-84 01:30:21 EDT
Article-I.D.: sri-arpa.1248
Posted: Mon Jun 25 01:30:21 1984
Date-Received: Thu, 28-Jun-84 03:56:20 EDT
Lines: 20

From:  Robert Amsler <AMSLER@SRI-AI.ARPA>

If one changed the content of a Spelling corrector to be a list of
predicates containing `facts' rather than sequences of letters, and then
one used such a program against the output of a parser which reduced
incoming text to similarly structured predicates, and the `fact checker'
then emitted confirmations or `corrections' of the facts in the parsed text
(e.g. South-Of San-Francisco San Jose; Capital-of USSR Moscow; etc.)
would this be a knowledge-based system? What has changed from sequences
of letters being acceptable `truths' to the mechanical use of predicates?

I fail to see how this is very different from having a spelling corrector
look over a string of letters and note that MAN and DOG are correct truths
whereas DOA (= Capital-of USSR San-Francisco) and MNA = (South-Of
San-Jose San-Francisco) are actually `misspellings' of DOG and MAN.

It might well be one doesn't want to call a system that uses this
strategy to proofcheck student's essays about geography an AI program,
but it sure would be hard to tell from its performance whether it
was an AI program or a non-AI program `pretending' to be an AI program.