The results showed that the words favoured most heavily by men were what grammarians call determinative words such as "the," "a," "as," "that" and "one." Female writers favoured "she" and relationship words such as "for," "with," "in," "and" and "not."
"This is surprising, since, unlike conversation, writing a book or an article does not involve direct social interaction"
Hmmm; if one wrote up such a program and applied it to, say, blogs on the web, I wonder what proportion it would sex accurately.
Update: the paper may be found here (though you have to subscribe to get the PDF). However, there is also a copy on the personal page of Prof. Moshe Koppel, one of the authors. And it appears that they're from Israel, not Illinois. (Perhaps the journalist confused the abbreviations?)
It appears that the paper may be found here:
http://www3.oup.co.uk/litlin/current/170401.sgm.abs.html
The abstract also mentions that the same technique may be used to determine whether a text is fiction or non-fiction.
I wish this program was running on a CGI so I could give it a go - it looks like fun. I wonder if it's capable of discriminating in technical writing, where your choice of words is much more restricted.