Humanist Discussion Group, Vol. 17, No. 419.
Centre for Computing in the Humanities, King's College London
www.kcl.ac.uk/humanities/cch/humanist/
www.princeton.edu/humanist/
Submit to: humanist@princeton.edu
Date: Tue, 02 Dec 2003 06:02:10 +0000
From: "Malcolm Hayward" <mhayward@iup.edu>
Subject: Re: 17.415 gender-testing fame: LLC in the Times
For those interested in gender testing in literature, I recently published
a study in the field: Malcolm Hayward, "Are Texts Recognizably Gendered? An
Experiment and Analysis." Poetics 31.2 (2003): 87-101. I took a different
approach than Moshe Koppel, Shlomo Argamon and Anat Rachel Shimoni. I
wanted to see if subjects (rather than a computer) could recognize
male/female authors. The idea here was that though there might be
differences in male/female prose, if the differences are unrecognized by a
reader, then that would seem to say something about both readers and
writers. I will reprint here the abstract:
"This paper describes an experiment testing the extent to which subjects
can determine an author's gender when passages of prose fiction are
presented anonymously. Forty 100 word passages, 20 each by male and female
authors, were presented to 47 subjects. Overall, subjects identified
gender correctly 61% of the time, indicating that writing may be, to some
degree, recognizably gendered. Female subjects, because of a bias to select
"male" as the author's gender, identified male writers correctly more often
than male subjects did. Certain significant differences were found between
the male and female subjects in the cues they claimed to use for their
decisions. Less experienced readers were found to be as successful as or
more successful than more experienced readers in identifying the author's
gender."
Note that the results are not as robust as the other study, 60% vs 80%, but
the passage length is smaller, and we are working here with humans, who
tend to get led down garden paths, make guesses based, perhaps, on extra
textual features (a tendency to guess "male"), and other undetected
elements. I'd interpret my study as more or less confirming the findings of
Koppel and the others.
Concerning the use of function words as markers, I'm not surprised, as
those working in the field of linguistic forensics, I guess it should be
called, such as John Burrows and David Holmes, have often found these to be
the best determiners of an author's identity. When I tried to figure out
what it was that led readers to correctly, or incorrectly, identify the
author's gender, the results were mixed at best. But I did discover
something interesting along the way. When examining the test passages, all
randomly selected, I found most of them contained dialogue. For the
passages by male writers, most of the dialogue was men talking to other
men. For the passages by female writers, the dialogue was preponderately
between men and women. I haven't yet gone back to test this further, but if
the results hold true--that guys write about guys talking to guys, whereas
gals write about girls talking to guys--that might explain some of the
differences in the use of function words (assuming, following Tannen and
others, that guy-talk differs from girl-talk and from girl-guy talk).
The more I worked on the topic, the more I realized how complex the
reading-writing-gender issue is. One might have guessed that genre would
have a role (detective novels a male realm? for example), but that did not
seem to work, and the case of Dick Francis that Koppel and the others
mention confirms that. So there is a lot left to be discovered about how
writers write gender--or are written by their gender--and how readers
interpret that writing.
Malcolm Hayward
This archive was generated by hypermail 2b30 : Tue Dec 02 2003 - 01:18:49 EST