"of the" need to be found, but will be filtered out afterwards by a rule that prevents phrases to end with a stopword.

What I'm currently doing is this (for 2 word phrases):

sub add_content {
    my $self = shift;
    my $content = shift;
    $words = [ split(/\s+/, $content) ];
        for ($i=0; $i < scalar(@$words) ; $i++) {
                my $first_word = lc($words->[$i]);
                my $second_word = lc($words->[$i+1]);

                # 2 word phrases
                if ($self->is_relevant_word($first_word , $second_word
+) && $first_word ne "$second_word") {
                        my $phrase = $first_word . " " . $second_word;
                        $self->{_related}{$phrase}++;
                        $self->_rate_phrase($phrase);
                }
    }
}
[download]

I'm just counting the occurences of phrases like this: $hash{$phrase}++ , and afterwards look for hash elements with values > 1.