The "eval" suggestion in the first reply didn't work for me, but the following did:

sub define_codes_method2 {
    my $language;

    while (<DATA>) {
        s/\s+$//;     # ("chomp" is too OS-dependent)
        next if !/\S/;

        if (/^%language=(.*)/) {
            $language = $1;
        }
        else {
            my ( $code, $descr ) = split /,/;
            $descr =~ s/\\x\{(\w{4})\}/chr(hex($1))/eg; # convert hex 
+values to chars
            $codes_hash{$code}{$language} = $descr;
        }
    }
}
[download]

Update: actually, the eval approach does work -- for some reason, when DATA has CRLF line-breaks and my unix "chomp" leaves the "\r" behind, nothing works. After fixing it so that the "s/.../chr(hex())/eg" approach would work, I tried the eval approach again, and that worked too:

       my ( $code, $descr ) = split /,/;
       $codes_hash{$code}{$language} = eval $descr;
[download]