This somewhat C-like solution is intended to avoid off-by-one errors at the beginning and end of the array. I have coded it as an array of array because that seemed logical and sufficient from the information so far -- the solution needs very little to make it a hash again. For example the operation chr(64+$_) would map the array indices in this AofA solution to the ascending alphabetical hash keys in the OP, should there be an undisclosed need for it to be a hash after all.

my @aOfA=();
for ( my $i=0; $i<=$#nlist; $i++ ) {
    my @tmp = ();
    ( $i > 0 ) and (($nlist[$i] - $nlist[$i-1]) < 2 )
      and $tmp[0]=$nlist[$i-1];
    push @tmp, $nlist[$i];
    ( $i < $#nlist ) and (($nlist[$i+1] - $nlist[$i]) < 2 )
      and push @tmp, $nlist[$i+1];
    push @aOfA, \@tmp;

    # Note that @tmp turns anonymous but doesn't dispose
    # at this reloop point
}
[download]