It is possible to do exactly what you asked for by explicitly testing for duplicates (use the function 'none' from List::Util) before storing. Because it uses arrays, it does preserve the order of the features. This solution is probably the slowest running of all the suggestion you received.

?type pearllearner315.pl
#!/usr/bin/perl
use strict;
use warnings;
use List::Util qw(none);
use Data::Dumper;
my $file = \<<'EOF';
bird beak
bird beak
bird claw
bird wings
bird feathers
snake fangs
snake scales
snake fangs
snake tail
EOF

#my $file = 'file.txt';


open( FILE, '<', $file ) or die $!;

my %hash;

while ( <FILE> ) {
    chomp;

    my $lines = $_;

    my ($key, $value) = split(/\s+/, $lines, 2);

    push @{ $hash{$key} }, $value if none {$value eq $_} @{$hash{$key}
+};
}
print Dumper(\%hash);


?perl pearllearner315.pl
$VAR1 = {
          'snake' => [
                       'fangs',
                       'scales',
                       'tail'
                     ],
          'bird' => [
                      'beak',
                      'claw',
                      'wings',
                      'feathers'
                    ]
        };

?
[download]