There isn't a general solution to this problem because of names (e.g, H.G. Wells) and quoting, but perhaps

/[.!?]\s{1,2}(?=[A-Z0-9])/
[download]

Update: Augmented regex for ! and ?

e.g. and i.e. should always be followed by a comma

Did you actually read the document you linked to? It clearly says:

Style guides do not agree on whether or not a comma should follow both these abbreviations ... The consensus seems to be in favor of the comma in American usage; against it in British usage.

Which is why I opened with a pedantic warning. It's true I was assuming he was parsing American English. And that's also why I gave a link, and not just my comment.

---

#11929 First ask yourself `How would I do this without a computer?' Then have the computer do it the same way.

> to be pedantic, e.g. and i.e. should always be followed by a comma

In this very sentence, none of them is followed by a comma.

($q=q:Sq=~/;[c](.)(.)/;chr(-||-|5+lengthSq)`"S|oS2"`map{chr |+ord
}map{substrSq`S_+|`|}3E|-|`7**2-3:)=~y+S|`+$1,++print+eval$q,q,a,
[download]

In general, for a corpus like this, I'd split it into known good, known bad, and grey, and then use test-driven development in order to build out my filter.

That's what I'm doing, but I got stuck at the early stage of handling just the cases of " e.g." and " i.e.", and I'm asking how to get unstuck so I can follow your advice, which I already was doing.

Did the negative look-ahead for a capital help? I should mention I think you have a typo in your real script (as opposed to what you posted) because the following script behaves well for me:

#!/usr/bin/perl
use strict;
use warnings;
use File::Stream;

my ($handler, $stream) = File::Stream->new(
    \*DATA,
    read_length => 1024,
    separator => qr/(?<!\b[A-Z])(?<!e\.g)(?<!i\.e)[.!?]\s{1,2}(?=[A-Z0
+-9])/,
  );


while (<$stream>) {
    print "*$_\n\n" ;
}



__DATA__
Perl filehandles are streams, but sometimes they just aren't powerful
enough.  This module offers to have streams from filehandles searched
with regexes and allows the global input record separator variable to
contain regexes.  Thus, readline() and the <> operator can now return
records delimited by regular expression matches.  There are some very
important gripes with applying regular expressions to (possibly
infinite) streams. Please read the CAVEATS section of this
documentation carfully.  Some bunnys are fluffy, e.g. Peter.  H.G.
Wells was a great author.  Some sports require specialized
equipment, e.g. baseball.
[download]

Debugging is hard without particular examples from your corpus.

---

#11929 First ask yourself `How would I do this without a computer?' Then have the computer do it the same way.

Any suggestions?

Yes: use Lingua::EN::Sentence.

I don't think  /(?<!( e|\.g)|( i|\.e))\.\s/ does what I think you think it does. If you have Perl version 5.10+ regex extensions, try something like (untested):

my ($exclude) =
  map  qr{ (?: \Q$_\E) (*SKIP) (*FAIL) }xms,
  join q{|},
  qw(e.g. i.e. Dr. Mr. Mrs. ... etc.)
  ;
my $delimiter = qr{ $exclude [.?!] \s }xms;
[download]

my ($exclude) =
  map qr{ (?: $_) (*SKIP) (*FAIL) }xms,
  join q{ | },
  map qq{\Q$_\E},
  reverse sort
  qw(e.g. i.e. Dr. Mr. Mrs. ... etc.)
  ;

my $delimiter = qr{ $exclude [.?!] \s }xms;
[download]

I have no idea how you could handle something like "H.G. Wells".

Update: I was a bit too quick with my post; see my update above. Also, I think I might see a way to exclude initialed names and similar things:

my $name = qr{ [[:upper:]] [[:lower:]]+ }xms;
my $initialed_name = qr{ \b [[:upper:]] [.] (?= \s+ $name) }xms;

my ($exclude) =
  map qr{ (?: $_) (*SKIP) (*FAIL) }xms,
  join q{ | },
  $initialed_name,
  map qq{\Q$_\E},
  reverse sort
  qw(e.g. i.e. Dr. Mr. Mrs. ... etc.)
  ;

my $delimiter = qr{ $exclude [.?!] \s }xms;
[download]

Update 2: It occurs to me that the above won't handle a name like P.D.Q. Bach, so maybe change  $initialed_name as follows (still untested):

my $initial = qr{ \b [[:upper:]] [.] \s*      }xms;
my $name    = qr{ \b [[:upper:]] [[:lower:]]* }xms;
my $initialed_name = qr{ $initial+ (?= \s+ $name) }xms;
[download]