From mboxrd@z Thu Jan  1 00:00:00 1970
Date: Thu, 11 Nov 1999 11:56:05 +0100
From: Elliott Hughes Elliott.Hughes@genedata.com
Subject: [9fans] Brian Kernighan?
Topicbox-Message-UUID: 9bd66c8c-eac8-11e9-9e20-41e7f4b1d025
Message-ID: <19991111105605.ZvNMk1khc8_Q2HthEgYNh9fmQcNEOq6px7trgE-sN-k@z>

forsyth wrote:
> the Cornell PL/1 compiler used a similar approach to
> do spelling correction on keywords as part of a broader
> attempt to repair all obvious errors in the given program; the results
> were amusing if not enlightening, as they so often are with AI.

IBM's jikes Java compiler also tries spelling correction, but its
ideas of proximity have nothing to do with any human's. it's
particularly unfortunate that it doesn't even know about the
Java naming conventions. (not that i necessarily think it should,
it's just that if it's going to try to guess what you meant to type,
it would be better off making educated guesses.)

there's a big difference between correcting simple typos and
more complicated "wrong identifier" errors.

anyway, back to the point: the original questioner might be
interested in "Finding Approximate Matches in Large Lexicons"
by Justin Zobel (jz@cs.rmit.oz.au) and Philip Dart (philip@cs.mu.oz.au),
which was the best paper i found when trying to come up with
decent guesses in a "dict"-like program.

btw, has anyone had better luck than i at getting information
about the CD-ROM version of the OED, with a view to having
a Plan 9/Unix OED "dict"? ever since i read the acme paper
with its "futtock" example, i've been jealous.

--
"As the Chinese say, 1001 words is worth more than a picture."
	-- John McCarthy