If you can advance incrementally through the text, you could try anchoring all of your patterns at pos with \G. Since pos is an lvalue, you could store the previous match position, try each pattern in priority order starting at the same previous position, take whichever match you prefer, store that into pos, and repeat for the next chunk. Something like: (untested)

my $lastpos = 0;
while ($lastpos < length $_) {
  my @matches = (undef x 3);
  pos = $lastpos;
  $matches[0] = pos if m/\G([^/?!"]+[.?!"])/gc;      #FIRST PRIORITY
  pos = $lastpos;
  $matches[1] = pos if m/\G([^:;-]+[:;-])/gc;      #SECOND PRIORITY
  pos = $lastpos;
  $matches[2] = pos if m/\G(.*(?:\n|\r|\z|$))/gc;   #LAST PRIORITY

  # somehow choose which match to use for the next cycle and set $last
+pos here
  #   substr $_, $lastpos, ($matches[$chosen] - $lastpos)
  #  should yield the selected chunk between choosing a match and upda
+ting $lastpos
}
[download]