Your script, as is, produced the expected results with my data. However, using either $` or $' is deprecated, in favor of ${^PREMATCH} and ${^POSTMATCH}, respectively--but you need to use the p option to preserve a copy of the matched string:

use strict;
use warnings;
use Data::Dumper;

my $token = 'perl';
my @contexts;
chomp( my @lines = <DATA> );

foreach my $line (@lines) {
    while ( $line =~ /\b $token \b/gip ) {

        my $before_string = ${^PREMATCH};
        my $after_string  = ${^POSTMATCH};

        if ( length $before_string > 5 ) {
            $before_string = substr( $before_string, -5 );
        }

        if ( length $after_string > 5 ) {
            $after_string = substr( $after_string, 0, 5 );
        }

        push @contexts, [ $before_string, $token, $after_string ];
    }
}

print Dumper \@contexts;

__DATA__
ABCDEFGHIJKLMNOP_PERL_ABCDCEFGHIJKLOMNP
0123456789ABCDEFGHIJKLMNOP_ PERL _ABCDCEFGHIJKLOMN_ PERL _P0123456789
ABCDEFGHIJKLMNOPPERLABCD PERL CEFGHIJKLOMNP
0123456789_ PERL _0123456789
[download]

Output:

$VAR1 = [
          [
            'MNOP_',
            'perl',
            '_ABCD'
          ],
          [
            'LOMN_',
            'perl',
            '_P012'
          ],
          [
            'LABCD',
            'perl',
            'CEFGH'
          ],
          [
            '6789_',
            'perl',
            '_0123'
          ]
        ];
[download]

Hope this helps!