Path: utzoo!attcan!uunet!spool2.mu.edu!sdd.hp.com!zaphod.mps.ohio-state.edu!pacific.mps.ohio-state.edu!linac!att!cbnewsl!npn
From: npn@cbnewsl.att.com (nils-peter.nelson)
Newsgroups: comp.text
Subject: Re: incorrect hyphenation in (old) troff
Summary: Unpredictability of hyphenation errors
Message-ID: <1991Jan24.223645.16630@cbnewsl.att.com>
Date: 24 Jan 91 22:36:45 GMT
References: <MJ7^8$@rpi.edu>
Organization: AT&T Bell Laboratories
Lines: 23

In article <MJ7^8$@rpi.edu>, fitz@mml0.meche.rpi.edu (Brian Fitzgerald) writes:
> Is there any way to predict incorrect hyphenation in troff?
> 
> How about a shell script (a little like spell) that compares a "stop
> list" to the text and perhaps prints warnings or generates ".hw"
> requests?
> 
> I am currently using the troff that came with SunOS 4.0.3.
> 
> Brian Fitzgerald


Sorry, but not so easy. The source code has a 19,000 byte table
of suffixes, and there is a binary encoded digram table.
Neither is user-accessible in the binary code.  We test the
algorithm by forcing the line length to something wee (.2 inches),
run a dictionary through it, and compare to a previously
hyphenated dictionary.
For DWB 3.2 we are adding the TeX hyphenation algorithm as a
user-selectable alternative. Seems to be more accurate, if
slower and larger, but may not be any more predictable.
If it's any consolation, either algorithm does better
than most people.