If it works on grammatical features I'd be wondering whether those assumptions are pre-programmed, or whether it refines its algorithm as it goes, in the light of the results it gets. If - which I doubt - there really is a noticeable gender difference in writing styles it would more insidiously problematic than some programmer with bunch of stereotypes in his/her head.