I suggest that a filter be added which strips the subject line of all non-alpha characters before scoring. It can be scored on the unstripped subject too, but on the stripped one too. That will detect messages where the spam words are broken up by dots, periods, dashes, etc.