By the way, depending on your system, /usr/dict/words may not be very useful at all for solving cryptograms. /usr/dict/words was originally intended for use with the spell utility, which does stemming to guess if a word is spelled correctly. Thus, /usr/dict/words may contain only root words and irregular forms, i.e. talk but not talks, talked, or talking.

I've been working on my own program in Perl to solve cryptograms. I recommend the ENABLE word list.

P.S. Oh, what the heck. My code works somewhat differently from merlyn's, so I'll go ahead and post it here. I expect his use of regexes will be faster than my use of split and hashes, though.

#!/usr/local/bin/perl -w

use strict;
use Getopt::Std;

use vars qw($opt_w);

getopts('w:') || die "Bad options.\n";

my $dict = $opt_w || 'wordlist';

open(DICT, $dict) or die "Can't open $dict: $!\n";

my $crypto = shift;

my $match = &crypto_canon($crypto);

while (<DICT>) {
    chomp;

    next if length $match != length $_;
    next if $match ne &crypto_canon($_);

    print "$_\n";
}

sub crypto_canon {
    my($word) = @_;

    my $canon;
    my %letters;
    my $next = 'a';

    foreach (split //, $word) {
        if ($_ !~ /[a-z]/) {
            $canon .= $_;
            next;
        } elsif (not exists $letters{$_}) {
            $letters{$_} = $next++;
        }
        $canon .= $letters{$_};
    }
    $canon;
}
[download]

1. Generate a unique combination of symbols. Don't repeat the same combination twice.
2. Decide whether or not the substitution generates meaningful output. This can either be done by manual inspection, or frequency analysis. If you choose to use frequency analysis, you'll need a decent set of frequencies to compare it against.

As far as your existing code is concerned, it might help if you put it in <CODE> tags.