Apr 27, 2007

Gender and Writing

Via Andrew Sullivan, this website uses an algorithm to predict whether text was written by a man or a woman.

I fed it three blog entries. The first was deemed overwhelmingly female, the second overwhelmingly male, and the third just barely female.

I gave it a piece on Sputnik and DARPA that I wrote for work, for which I was determined to be very male. Then I gave it a paragraph on the health risks of trans fats, which was apparently very female. A term paper on neoconservatism was slightly female. Another term paper, on intelligence reform, was way, way male.

Looking at the keywords, it seems that the algorithm is designed to detect gendered topics rather than the sex of the writer. Or perhaps to detect the sex of the writer through the gender of the topic. Pronouns like I, me, your, she, hers, we, and myself are supposedly female--so if you're writing mostly about yourself and other people, you're a girl. (Livejournal is overwhelmingly female, I hear.) So is "was," which might be used to talk about recent events. Prepositions like "with," "where" and "when" are female, too. (E.g., "I was with so-and-so when we ran into so-and-so in the bar where I met so-and-so.")

The male keywords contain more spatial prepositions (below, above, at, to, around) and present forms of be (are, is). Non-gendered and indefinite pronouns--like it, what, who, these, and many--are male as well. Articles (a, the) too. So if you write about objects and the concrete realities of the world as it is now, you're a guy.

I may be completely wrong, but suspect that if this algorithm works on most people, it's because of topics. You're not going to find much room for "I" "me," "hers," or "myself" in a paper about intelligence reform.

---

I think that something profoundly stupid will happen this weekend. Maybe several somethings.

No comments: