Nice! I came up with: perl -nle '$w{lc s/[[:punct:]]//r}++ foreach split(/\s+/,$_); print map{"$w{$_} $_ \n"} (sort{$w{$a} <=> $w{$b}} keys(%w)) if eof()' < k

Nice! I came up with: perl -nle '$w{lc s/[[:punct:]]//r}++ foreach split(/\s+/,$_); print map{"$w{$_} $_ \n"} (sort{$w{$a} <=> $w{$b}} keys(%w)) if eof()' < kjvbible.txt If anyone's curious this takes .47 seconds on my local machine whereas the optimized shell solution takes about 3.5 seconds. Edit: It looks like mhd's usage of autosplit -a and @F are faster than split($_) but calling print foreach key is slower than my print map strategy. So our solutions are combined to get: perl -anle '$w{lc s/[[:punct:]]//r}++ foreach @F; END{print map{"$w{$_} $_ \n"} (sort{$w{$a} <=> $w{$b}} keys(%w))}' < kjvbible.txt Which runs in .39 seconds. mhd's solution runs in .42 seconds. For reference the optimized C version takes .059 seconds (possibly cause of clang?), wc -w takes .023s and the optimized python version takes .288