I agree with saintmike. What you want to do is better using OO. To use the approach you are now, you can do something like this. See my deepmerge function. It is not complete, but works for your sample data. To complete it you need to handle all data types.

#!/usr/bin/perl
use strict;
use Data::Dumper;

my %europe = (
        skin => "euro-white",
        language => {
                default => "euro-englisch",
                instrument => "euro-clarinet",
        },
        politics => "euro-democratic",
);
my %german = (
        skin => "de-white",
        language => {
                default => "de-german"
        }
);
my %bavarian = (
        skin => "bav-white",
        language => {
                default => "bav-bavarian"
        }
);

# this is not enough, german/language overwrites
# the complete hash of euro/language
my %woman = %europe;

deepmerge(\%woman, \%german);
deepmerge(\%woman, \%bavarian);

print Dumper(\%woman);
# result is this:
# $VAR1 = {
#           'skin' => 'bav-white',
#           'politics' => 'euro-democratic',
#           'language' => {
#                           'default' => 'bav-bavarian'
# ** this is not here: instrument => "euro-clarinet",
#                         }
#         };
unless ($woman{language}->{instrument}) {
        die "Error: Language instrument is not there";
} else {
        print "It worked\n";
}

sub deepmerge {
    my ($into, $this) = @_;
    
    for my $k (keys %$this) {
        if (defined $k and 
            exists $into->{$k} and 
            defined $into->{$k} and 
            ref($into->{$k}) eq 'HASH' and
            ref($this->{$k}) eq 'HASH') {
        
            deepmerge($into->{$k}, $this->{$k});
        
        } elsif (defined $k) {
            $into->{$k} = $this->{$k};
        }
    }
}
[download]