Path: utzoo!mnetor!uunet!lll-winken!lll-lcc!ames!umd5!purdue!i.cc.purdue.edu!j.cc.purdue.edu!pur-ee!uiucdcs!uiucdcsb!kadie
From: kadie@uiucdcsb.cs.uiuc.edu
Newsgroups: comp.ai
Subject: Grammar Checking
Message-ID: <165000007@uiucdcsb>
Date: 21 Jan 88 02:50:00 GMT
Lines: 30
Nf-ID: #N:uiucdcsb:165000007:000:1081
Nf-From: uiucdcsb.cs.uiuc.edu!kadie    Jan 20 20:50:00 1988



Remember last year's heated discussion about grammar and
style checkers? Well here is a little data (some few data?).

I recently had a 13 page double-spaced document proof read
by a person (my advisor). He suggested about 22 simple grammar corrections;
I made every correction. Then I feed the same document to RIGHTWRITER
a commercial program for the IBM AT. It make 159 suggestions; 
I took 9 of them. The person and the program make one identical
suggestion.

So:
* Humans are much better proofreaders than (today's) programs.
* Program might still be worth using since they may find errors
  that a human misses and since they are convenient.


Also:
The most important comments a human makes are about the understandability
of the document. In a sense the human is telling how well
the document "executes."  Since, the program only looks a syntax,
it can only guess about this part.


Carl Kadie
Inductive Learning Group
University of Illinois at Urbana-Champaign
UUCP: {ihnp4,pur-ee,convex}!uiucdcs!kadie
CSNET: kadie@UIUC.CSNET
ARPA: kadie@M.CS.UIUC.EDU (kadie@UIUC.ARPA)