/\s*(?=\w+\s+\w+)[\w\s]{49,}\w/
[download]

lima1 wants minimal matches, so the {49,} should actually be {49,}?, in which case it stops working.

The reason is that the look-ahead is not limited to what the [\w\s]{49,}? matches. A small demonstration:

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

my $re = qr{^\s*(?=\w+\s+\w+)[\w\s]{49,}?\w};

my $str =  ('x' x 65) . ' x';

if ($str =~ m/$re/) {
    print $&, $/;
}
__END__
xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
[download]

You see that match only includes a long word, not a sentence. There is a sentence present, but it's not matched.

Update: In Perl 6 you could use & like this:

regex short_word {
   \s* 
   [ 
   & \w+ \s+ \w+ 
   & .**?{1..50}
   ] 
}
[download]

Well, I am not a native English speaker, but as I understand it...

I want to match sentences (everything that is not just a long word) of a minimum length

...doesn't mean that the length of the sentence has to be minimal but that the length of the sentence has to be equal or bigger than a $minimum_length.

Anyway, if a minimal match is what you want, it still can be done with a regexp (without embedded code!), though a complicated one:

sub make_re {
    my $len = shift;
    $len > 3 or die "len <= 3";
    my $re = "\\b"
             . join('|', map("\\w(?:\\s[\\s\\w]{$_,}?", reverse 1 .. (
+$len - 3)), "\\w+\\s+")
             . (")" x ($len - 3)) . "\\w+";
    warn "re: /$re/\n";
    return qr/($re)/;
}

my $re = make_re(5);
while(<DATA>) {
    print "$1\n" if $_ =~ $re;
}
__DATA__
foo foo
foooooooo
foooooooo fooo
foo foo foo foo foo
foo foooooooo foo foo
foo f foo
fo fo foo
f fo foo
f fo
[download]

Ah, nice. Seems to work perfectly! Thank you very much...

Update:

++moritz. But that's still ok for me, because it filters my problematic case (just a long word). That's all I need here.

m/\A \s* (\w+\s+\w+?.*?) (??{ length($1) >= 50 }) \s* \z/xs;
[download]

(Update: this regex might be very inefficient because both \w+? and .*? might match the same number characters. Use (\w+\s+\w.*?) instead.)

In general that sounds like a problem for which regexes aren't the best solution (split and friends might be better).

Thanks! Yeah, this was also the only way I saw, but I was scared by the experimental warning in perlre. I need a regex here because the API I use requires a regex.