Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!utgpu!water!watnot!ccplumb
From: ccplumb@watnot.UUCP
Newsgroups: comp.unix.wizards,comp.unix.questions
Subject: Re: Spell and /usr/dict/web2
Message-ID: <12630@watnot.UUCP>
Date: Wed, 18-Mar-87 10:29:10 EST
Article-I.D.: watnot.12630
Posted: Wed Mar 18 10:29:10 1987
Date-Received: Fri, 20-Mar-87 06:20:54 EST
References: <135@utecfb.Toronto.Edu> <2563@hcr.UUCP>
Reply-To: ccplumb@watnot.UUCP (Colin Plumb)
Organization: U. of Waterloo, Ontario
Lines: 25
Keywords: spell dictionary web2
Xref: utgpu comp.unix.wizards:1439 comp.unix.questions:1421

In article <2563@hcr.UUCP> mike@hcr.UUCP (Mike Tilson) writes:
>The question was whether one would get a better "spell" hash table
>by building it from the Webster's word list rather than from the
>much shorter dictionary used by the "spell" makefile.
>
>I believe the answer is "no".
>
>First, if you do greatly expand the dictionary, remember to increase
>the size of the hash table.  The algorithm works best when the final
>bit table has about 50% ones and 50% zeros.  With a larger dictionary
>and no increase in table size, you will turn on more bits.  In the limit,
>all bits will be on, and all input to spell will be accepted as correct
>spelling.

Um... not quite.  If the table is 50% ones, then 50% of any collection of
random character strings would be accepted as correct.
As I recall, from Jon Bentley's discussion of spell in Programming Pearls,
the hash table size was chosen so that, assuming 20 reported problems per
run, a misspelling would slip through about once every hundred runs, so
the ones should fill up 1/2000 of the hash table.
--
	-Colin Plumb (watmath!watnot!ccplumb)

Zippy says:
My CODE of ETHICS is vacationing at famed SCHROON LAKE in upstate New York!!