heh although that would work I think sticking with natural language is better, it's really not that hard to parse "word word number word number"