Tendo City

This is far from new, but I just came across it.

Odd concept...

The idea:
http://www.nytimes.com/2003/08/10/magazi...82&ei=5070
http://www.nytimes.com/2003/08/10/magazi...9a&ei=5070

Application:
http://bookblog.net/gender/genie.php

Tried on this post (choosing a long one of mine at random):
http://www.tcforums.com/forums/showpost....ostcount=2

Results (full post): 3743 male/ 2498 female, thinks the author is male.
Results (just the two long quotes): 2601 male, 1511 female, thinks author is male.
Results (just the new unquoted part at the bottom): 961 male, 849 female, thinks the author is male.

Interesting? I'm not sure, the thesis seems pretty questionable. Sure in some cases gender may seem obvious, but there's no way this would work as well as they suggest... it's going to be wrong quite a bit, I'm sure.

What evidence do they have to back this up, and why do they pick these keywords? Kinda strange if you as me... I should think they'd have some better results than cherry picking from data where they already know the results... If true though, it certainly would be something interesting I suppose. I just don't see it as having much more evidence than the poor case of those "semi emotions" that some people tried to foist on us years back. Anyway, I guess what I have to say is where's the evidence, and with what methods did they get it? That's information that could convince me.

Oh, I see when this was written, in 2003. ABF, do you have any more current info on this?

Well, according to the site the Gender Genie thing itself has been updated as recently as last year. That doesn't mean it actually works well, though. As I said, the idea seems pretty specious...

http://www.slumdance.com/blogs/brian_fle...00327.html



Evidently the guy who wrote it doesn't think it's too accurate either...

http://www.mentalfloss.com/blogs/archives/6501

Quote: Most of the time, people drop their writing into [The Gender Genie] and, when they don’t get the result they expect, declare it to be wrong, wrong, wrong. Yet, a lot of its users still find it and its analysis to be a fun time waster. Despite having written the program, I didn’t come up with the algorithm and believe that the Genie works no better than the flip of a coin. However, I don’t think it to be a complete time waster since there actually is some academic study that went into it.

In the most basic terms, the computational linguists behind the algorithm, Koppel and Argamon, took a bunch of fiction and looked for trends based on gender. Using complicated formulas, they determined that male writers tended to write more about specific things like an apple, a book, or the car. In contrast, female writers wrote about connections to things like my apple, your book, or our car. The nouns themselves (apple, book, car) didn’t matter much but the preceding qualifier, whether an article (a, an, the) or possessive (my, your, our), did.

I see, so at the very least they learned that their method isn't very good.

I figured as much. Punching in my posts so far I've been getting half and half, and most of the time it's pretty close. Really short posts seem to get no "hits" in female it seems. I checked their list though. It's not like we'd be using a number of those words anyway.