Thank you all for your replies, here are the scripts, data and other pertinent information as requested. My Perl version is 5.36.0 on Linux and the sample text file I used in the following "benchmarks" was created with:

wget http://www.astro.sunysb.edu/fwalter/AST389/TEXTS/Nightfall.htm
html2text-cpp Nightfall.htm >nightfall.txt
for i in {1..1000}; do cat nightfall.txt >>in.txt; done
[download]

Now, in.txt is 77MB large. Bash/sed script:

#!/bin/bash

cat *.txt | \
    tr -d '[:punct:]' | \
    sed 's/[0-9]//g' | \
    sed 's/w\(as\|ere\)/be/gi' | \
    sed 's/ need.* / need /gi' | \
    sed 's/ .*meant.* / mean /gi' | \
    sed 's/ .*work.* / work /gi' | \
    sed 's/ .*read.* / read /gi' | \
    sed 's/ .*allow.* / allow /gi' | \
    sed 's/ .*gave.* / give /gi' | \
    sed 's/ .*bought.* / buy /gi' | \
    sed 's/ .*want.* / want /gi' | \
    sed 's/ .*hear.* / hear /gi' | \
    sed 's/ .*came.* / come /gi' | \
    sed 's/ .*destr.* / destroy /gi' | \
    sed 's/ .*paid.* / pay /gi' | \
    sed 's/ .*selve.* / self /gi' | \
    sed 's/ .*self.* / self /gi' | \
    sed 's/ .*cities.* / city /gi' | \
    sed 's/ .*fight.* / fight /gi' | \
    sed 's/ .*creat.* / create /gi' | \
    sed 's/ .*makin.* / make /gi' | \
    sed 's/ .*includ.* / include /gi' | \
    sed 's/ .*mean.* / mean /gi' | \
    sed 's/ talk.* / talk /gi' | \
    sed 's/ going / go /gi' | \
    sed 's/ getting / get /gi' | \
    sed 's/ start.* / start /gi' | \
    sed 's/ goes / go /gi' | \
    sed 's/ knew / know /gi' | \
    sed 's/ trying / try /gi' | \
    sed 's/ tried / try /gi' | \
    sed 's/ told / tell /gi' | \
    sed 's/ coming / come /gi' | \
    sed 's/ saying / say /gi' | \
    sed 's/ men / man /gi' | \
    sed 's/ women / woman /gi' | \
    sed 's/ took / take /gi' | \
    sed 's/ tak.* / take /gi' | \
    sed 's/ lying / lie /gi' | \
    sed 's/ dying / die /gi' | \
    sed 's/ made /make /gi' | \
    sed 's/ used.* / use /gi' | \
    sed 's/ using.* / use /gi' \
>|out-sed.dat
[download]

This script executes in around 5 seconds:

% time ./re.sh 
real    0m5,201s
user    0m43,394s
sys    0m1,302s
[download]

First Perl script, slurping input file at once and processing line-by-line:

#!/usr/bin/perl
use strict;
use warnings;
use 5.36.0;

my $BLOCKSIZE = 1024 * 1024 * 128;
my $data;
my $IN;
my $out='out-perl.dat';
truncate $out, 0;
open my $OUT, '>>', $out;

my @text = glob("*.txt");
foreach my $t (@text) {
  open($IN, '<', $t) or next;
  read($IN, $data, $BLOCKSIZE);
  my @line = split /\n/, $data;
  foreach (@line) {
    s/[[:punct:]]/ /g;
    tr/[0-9]//d;
    s/w(as|ere)/be/gi;
    s/\sneed.*/ need /gi;
    s/\s.*meant.*/ mean /gi;
    s/\s.*work.*/ work /gi;
    s/\s.*read.*/ read /gi;
    s/\s.*allow.*/ allow /gi;
    s/\s.*gave.*/ give /gi;
    s/\s.*bought.*/ buy /gi;
    s/\s.*want.*/ want /gi;
    s/\s.*hear.*/ hear /gi;
    s/\s.*came.*/ come /gi;
    s/\s.*destr.*/ destroy /gi;
    s/\s.*paid.*/ pay /gi;
    s/\s.*selve.*/ self /gi;
    s/\s.*self.*/ self /gi;
    s/\s.*cities.*/ city /gi;
    s/\s.*fight.*/ fight /gi;
    s/\s.*creat.*/ create /gi;
    s/\s.*makin.*/ make /gi;
    s/\s.*includ.*/ include /gi;
    s/\s.*mean.*/ mean /gi;
    s/\stalk.*/ talk /gi;
    s/\sgoing / go /gi;
    s/\sgetting / get /gi;
    s/\sstart.*/ start /gi;
    s/\sgoes / go /gi;
    s/\sknew / know /gi;
    s/\strying / try /gi;
    s/\stried / try /gi;
    s/\stold / tell /gi;
    s/\scoming / come /gi;
    s/\ssaying / say /gi;
    s/\smen / man /gi;
    s/\swomen / woman /gi;
    s/\stook / take /gi;
    s/\stak.*/ take /gi;
    s/\slying / lie /gi;
    s/\sdying / die /gi;
    s/\smade /make /gi;
    s/\sused.*/ use /gi;
    s/\susing.*/ use /gi;

    close $IN;
    print $OUT "$_\n";
  }
}
[download]

Please, ignore the technicality of failed matches before/after a newline, as this line-by-line implementation is uselessly slow anyway at over 4 minutes. Time to slurp the input and split it in lines < 1 second.

% time ./re1.pl 
real    4m1,655s
user    4m29,242s
sys    0m0,380s
[download]

If I split by /\s/ instead, it consumes 5 seconds at it, but the substitutions take 1 minute, i.e. 12 times slower than bash/sed:

% time ./re2.pl 
real    1m5,096s
user    1m11,889s
sys    0m0,524s
[download]

Final test, I created 1000 copies of nightfall.txt (77KB) with % for i in {1..1000}; do cp nightfall.txt nightfall-$i.txt; done. All scripts took roughly the same amount of time to complete. So, it would seem that my initial estimation of "60-70% slower Perl" was very optimistic, as the full scripts perform other tasks too, where Perl's operators and conditionals obviously blow Bash's out of the water.

For the record, I do all file read/write operations on tmpfs (ramdisk), so disk I/O isn't an issue. I'll implement AnomalousMonk's solution with hash lookup and report back soonest.

An idea that just occured to me is that when doing matches in word-splits, most regexes can apparently terminate the loop (and next;) as no further matches are expected below. Still, I'd like to exhaust all possibilities before admitting defeat.

---

---

• Are you posting in the right place? Check out Where do I post X? to know for sure.
• Posts may use any of the Perl Monks Approved HTML tags. Currently these include the following:

  <code> <a> <b> <big> <blockquote> <br /> <dd> <dl> <dt> <em> <font> <h1> <h2> <h3> <h4> <h5> <h6> <hr /> <i> <li> <nbsp> <ol> <p> <small> <strike> <strong> <sub> <sup> <table> <td> <th> <tr> <tt> <u> <ul>

• Snippets of code should be wrapped in <code> tags not <pre> tags. In fact, <pre> tags should generally be avoided. If they must be used, extreme care should be taken to ensure that their contents do not have long lines (<70 chars), in order to prevent horizontal scrolling (and possible janitor intervention).
• Want more info? How to link or How to display code and escape characters are good places to start.