THE INTEGRATION OF BANKING AND TELECOMMUNICATIONS: THE NEED FOR REGULATORY REFORM

BEST PRACTICES 365

was repeatedly not selected by a blind testing of stylemarker
comparison.
One argument made against my prior work is that the
stylemarkers were tested independently and not combined, but it
is supposedly the combination of an unknown number of
stylemarkers that supports the contention that each person has a
unique authorial style.^77 However, anyone reading the test results
could combine them, and when combined, the accuracy rate at
identifying a questioned document to the real author in a pool of
four authors for a combination of forensic stylistics stylemarkers
is about fifty-two percent.
Forensic stylistics has very poor accuracy on ground-truth
data where no one is preselected as author prior to K/Q feature
selection. It is not a reliable method for authorship
identification. The poor reliability of forensic stylistics, as
reported in my prior article,^78 was later confirmed by validation
testing using different ground-truth data by St. Vincent and
Hamilton,^79 Koppel and Schler,^80 and Chaski.^81

C. No Relationship to Standard Linguistic Methodology

Crystal^82 provided a surprisingly caustic but accurate review
of McMenamin.^83

M[cMenamin] talks in a semistatistical way (“It is

extremely unlikely that this close lexical match in

profanity could be due to chance coincidence... .”) but

he does not present the statistical analysis which would

make such comparisons convincing. Indeed, at several

points, one wonders whether it would in principle be

(^77) See id.
(^78) See id. at 3.
(^79) See S. St. Vincent & T. Hamilton, Author Identification with Simple
Statistical Methods, SWARTHMORE COLL., DEP’T OF COMPUTER SCI. (2001)
(on file with author).
(^80) Koppel & Schler, supra note 3.
(^81) Chaski, Empirically Testing, supra note 16.
(^82) David Crystal, Book Review, 71 LANGUAGE 381, 381–85 (1995)
(reviewing MCMENAMIN, supra note 2).
(^83) MCMENAMIN, supra note 2.