$punc = "?!.";
ucfirst($string);
$string =~ s#([$punc][^a-z]*[a-z])#uc($1)#eg;
## or 
$string =~ s#([$punc]\s*[a-z]#uc($1)#eg;
[download]

The first meets the strict definition of the requirements, but also causes
"Who are you? 6 is my number"
to become
"Who are you? 6 Is my number"
which may or may not be desired. If not, the second one only allows whitespace to be skipped between the end of a sentence and the start of a word.

I used a-z instead of \w because otherwise perl will try to "uppercase" an underscore, which is not what we want. It's also safe to put the whole thing in a single paren, and to uppercase the whole thing, since the only part able to be uppercased is the part we *want* uppercased. I also removed the specification of what constitutes the end of a sentence outside of the regular expression, which makes it a little easier to read, and also allows easier changes in the future without accidentally breaking your regular expression.

The usual warning vis a vis use English; and \w apply, of course. :)

my $d=ucfirst("sample. data some! will be. captilized");
$d=$s/([\!\.]\s*)(\w)/$1\U$2\U/g;
[download]

while (<>)
{
  $_ = ucfirst;
  s/([!.?])(\s+)(\w)/$1$2\U$3/g;

  print;
}
[download]

$string =~ s/((?:^\s*|[!?.]+\s+)[^a-zA-Z0-9]*)([a-z])/$1\u$2/g;
[download]