Random: Snort of the Medusa

Hurray! The Gender Genie is back: that online widget into which you paste a selection of text, indicate whether it's fiction or non-fiction or, in this incarnation, a blog entry, and it tries to figure out whether the author is male or female. I wonder what kind of tweaking they've done to the algorithm to make "blog entry" a category to itself. The paper on the original algorithm by Moshe Koppel (Bar-Ilan University) and Shlomo Argamon (Illinois Institute of Technology) is over here, but I haven't had a chance to read it through yet. The Gender Genie people claim only that their widget "uses a simplified version" of the Koppel-Argamon test.

It's all about keywords, of course; in judging fiction, it looks like they think the subjunctive mood is feminine and the indicative mood is masculine; personal pronouns are feminine while relative pronouns are masculine; and, in a final twist of oddness, the simple past of the verb "to be" is feminine while the present tense is masculine. (Um, sure.)

So of course I tested a few of my own pieces: all the lengthy ones, because the more words, the better. "Ubi Sunt Gaudia", written from Ed's POV, splits almost 50-50 on keywords, with a tilt to the masculine side. I'm a boy! "Light the Traveler Home" (unrevised version), from Winry's POV, again splits quite close to fifty-fifty, but with a slightly more emphatic tilt toward the feminine. I'm a girl! "Forsan Et Haec", about Maria Ross, ditto. Still a girl! "Tick-Tock", Ed's POV, back to leaning slightly masculine. Boy! "Ivory Gate", split POV between Ed and Noa, a somewhat more substantive lean feminine. Girl! "Country Matters", Pinako's POV, a close-run thing, but in the end it just tips over into the feminine. Girl! Finally, a diptych of radically different styles: "A Story To Tell" (an action adventure in which Ed and Al stop some drug smugglers) and "Passio", a short in which Winry and Al sit in a hospital waiting room while Ed undergoes surgery. The former genies out firmly male and the latter wildly female.

So, taken all for all, it looks like this widget is very good about sussing out the narrative POV of fiction (or I'm pretty good at just shifting my language enough to make male narrators sound male and female narrators sound female, particularly under emotional stress), but less good about actually determining the gender of the author. I'll try some of my non-fiction later -- see what it makes of my dissertation, maybe ...

UPDATE: By the way, this blog entry? Undeniably masculine, by a 2-to-1 keyword margin. I'm a boy!