You're trying to do it with a character class, which doesn't act as you want. Try matching all text up to what you capture and substituting just the capture.

If you're removing so much of the data from the string, why not bin it and make a new one. Something like:-

  my @out = ();
  while( m/(book|cover)/g ) {
    push(@out, $1);
  }
  my $newstr = join(" ", @out);
[download]

#!/usr/bin/perl
use strict;
use warnings;

my $zm = "ble ble book uff uff cover asdf asdfa ala book coverr";
$zm =  join(' ', grep {/\b(book|cover)\b/} (split / /,$zm));
print $zm;
[download]

Check out the following Perl documentation: perlretut, perlre and perlrequick.

s/[^bc]+      # Any characters that aren't a 'b' or an 'c'.
 |b(?!ook)    # Or a 'b' which doesn't start a 'book'.
 |c(?!over)   # Or a 'c' which doesn't start a 'cover'.
 /gsx;
[download]