It seems to me that the sticking point is going to be deciding what qualifies as "pure text." Getting the content out of the XML is fairly trivial: just walk recursively through the XML file after loading it into some XML parsing module, and grab the values of the "content" keys. (I only downloaded about 0.2% of the file as a sample, but that appears to be consistent.) The simple bit of code below does that, counts the "words" in a dictionary hash, and outputs the sorted results. However, since it splits the text on whitespace, the resulting words contain a lot of punctuation, including wiki formatting. So you'll have to parse that out, and also deal with other issues: Unicode and HTML encoded characters, embedded HTML tags, "wide characters," and more.

#!/usr/bin/env perl
use Modern::Perl;
use XML::Simple;
use Data::Dumper;

my $xml = XML::Simple->new();
my $in = $xml->XMLin('wiki.xml');
my %dict;
walk($in);

for (sort {$a cmp $b} keys %dict){
    say "$dict{$_}    $_";
}

sub walk {
    my $h = shift;
    for my $k (keys %$h){
        if($k eq 'content'){
        add_to_dict($h->{$k});
    } elsif( ref($h->{$k}) eq 'HASH' ){
        walk($h->{$k});
        }
    }
}

sub add_to_dict {
    my $text = shift;
    for my $w (split /\s+/, $text){
    $dict{$w}++;
    }
}
[download]