Wouldn't it be easier to write it (and read it) this way?:

use strict;

use XML::Parser;
use XML::XPath;
use Lingua::StopWords qw( getStopWords );

my $stopwords = getStopWords('en');

my $file = $ARGV[0];

my $xp = XML::XPath->new(filename=>$file);

for my $n(1..600)
{
    my %seent;
    my $textnodeset = $xp->find('//pair[@id = '.$n.']/tAnnotation/tree
+/node/word/attribute[@name="token"]');
    my @texts = map $_->string_value, $textnodeset->get_nodelist;
    my @termst = grep{!$seent{$_}++ && !$stopwords->{$_}} @texts;
        
    my %seenh;
    my $hyponodeset = $xp->find('//pair[@id = '.$n.']/hAnnotation/tree
+/node/word/attribute[@name="token"]');
    my @hypos = map $_->string_value, $hyponodeset->get_nodelist;
    my @termsh = grep{!$seenh{$_}++ && !$stopwords->{$_}} @hypos;
    
    for my $i (@termst)
    {
        for my $j (@termsh)
            { print "$i $j\n" }
    }
}
[download]

You could search google for 'xpath unique', so that xpath returns only unique values. But I am not sure if it will speed things up or slow them down...