Basically, you have to take whitespace-separated words, removing non-alphanumeric (i.e. punctation) characters from the beginning or the end. Punctation is allowed inside a word. What makes this a bit more complicated is that in some languages, the space os omitted around em dashes, so you have to consider em-dashes as whitespace too. Thus, it's easier to define what separates words than what counts as a word. Here's a simple solution.

#!perl

use warnings;
use strict;

my # I hate to say this, as it's not mine
$text = q{

"'My Dear Mr. Sherlock Holmes:--"Lord Backwater tells me that I may
place implicit reliance upon your judgement and discretion. I have
determined, therefore, to call upon you and to consult you in referenc
+e
to the very painful event which has occurred in connection with my
wedding. Mr. Lestrade, of Scotland Yard, is acting already in the matt
+er,
but he assures me that he sees no objection to your co-operation,
and that he even thinks that it might be of some assistance. I will
call at four o'clock in the afternoon, and, should you have any other
engagement at that time, I hope that you will postpone it, as this
matter is of paramount importance. Yours faithfully, ST. SIMON.'

 -- Sir Arthur Conan Doyle, The Adventures of Sherlock Holmes. 1892.

};

use locale;
my $stuff_between_words = qr{
  [^[:alnum:][:space:]]*
  (?: [[:space:]]+ | --+ | ^ )
  [^[:alnum:]]*
 }x;
my @words = split $stuff_between_words, $text;
shift @words; # remove the empty word from the beginning

print join("  ", map "<<$_>>", @words), "\n";

__END__
[download]

However, if I'd really need a fast solution, I'd use flex.

Update: on second thought, it'd be better to use just -- instead of --+.