Sorry Aristotle. Fletch's (partial) solution, neat as the technique is, falls foul of the fact that deleting the apostrophies in a one range, causes all the subsequent columns to shift.

I should have tested. Anyway, in this case, it's a simple matter of changing the order of operations:

{
    local *_ = \substr $source, $start, $len;
     y/a-zA-Z0-9\n\|-/ /c;
     y/'//d;
}
[download]

However, that obviously only works if there's only one operation affecting length. For a more general case, I'd do something like this (untested):

#!/usr/bin/perl -w
use strict;

my @range =
    map /^(\d+)-(\d+)$/,
    sort { $a <=> $b }
    splice @ARGV, 1;

unshift @range, 0;
$range[$_] = 1 + $range[$_+1] - $range[$_] for 0 .. $#range-1;
$range[-1] = '*';

die "Negative length field specified"
    if grep $_ < 0, @range[0 .. $#range-1];

my $fmt = join " ", map "A$_", @range;

# pick odd numbered elements
my @selected = map 1 + $_ * 2, 0..$range_/2;

while(<>) {
    my @field = unpack $fmt, $_;
    for (@field[@selected]) {
        tr/a-zA-Z0-9\n\|\-'/ /c;
        tr/'//d;
    }
    print join '', @field;
}
[download]

The point is to structure your data whenever possible. An array element end is never ambiguous, a \x7F can happen to be, and in my case, whatever my mark character, I've always been bitten by it.

Makeshifts last the longest.