338 JOURNAL OF LAW AND POLICY
exemplars are typically not edited to any conventional,
newspaper, academic, or industrial standards. If the researcher
is not using a forensically feasible dataset to test his method, he
might be misled into thinking that his method—built to assign
clean, grammatical, edited business letters, newspaper articles,
or novels—will work accurately on messy, ungrammatical,
forensically significant texts. Essentially, building a method
without testing it on forensically feasible data simply
overgeneralizes a method’s ability: it may look scientific because
there are some validation tests to refer to, but the validation test
results do not prove that the method can work on the data in the
case or any forensically feasible data.
Research that focuses on literary classics or edited
newspaper articles may develop accurate methods, but these
methods must be tested on forensically feasible data before they
are borrowed across-the-board for forensic authorship
identification. In most cases, literary methods fail to work on
forensic data simply because the literacy methods require far
longer texts than the forensic case affords. Brevity is a fact of
life inherent in forensic authorship identification that cannot be
avoided or helped by research that focuses on texts that contains
thousands, tens of thousands, or hundreds of thousands of
words. Again, using methods that work well on literary texts or
newspaper text banks, without independently testing the methods
on forensically feasible data, may appear to be scientific because
there is published literature in humanities computing to refer to
about authorship identification in nonforensic settings,^14 but
using such methods is akin to using a screwdriver on a nail—and
an unvalidated screwdriver at that.
the shift key, producing a typical typing error for novices, while the
handwriter never made that kind of mechanical error. The context of this
difference was not noted in the forensic stylistics report by Agent Fitzgerald;
instead he argued that [don;t] was a unique stylemarker.
(^14) For instance, the Association for Computers and the Humanities
publishes Literary and Linguistic Computing, a journal where authorship
issues in literature, religion and other nonforensic settings are regularly
discussed.