Robotics Reference
In-Depth Information
them was written by one of two different people: Alexander Hamilton, a
Federalist leader, and James Madison, the fourth president of the U.S.A.
Because of this uncertainty of authorship the essays became known as the
“disputed papers”.
23
Mosteller and Wallace measured the frequencies of the so-called func-
tion words—prepositions, conjunctions and both definite and indefinite
articles—in the known writings of Hamilton and Madison. For exam-
ple, they found that the word “upon” appeared 3.24 times per 1,000
words in the writings of Hamilton but only 0.23 times per 1,000 in
those of Madison. On the basis of these frequency measures Mosteller
and Wallace concluded that all 12 of the disputed papers were written by
Madison, a conclusion that agreed with the vast majority of those history
scholars who had discussed the papers' authorships.
More recently, stylometry has benefited from the application of Arti-
ficial Intelligence techniques. Neural networks have the ability to recog-
nize the underlying organization of a set of data and therefore can be
trained to learn to differentiate between the writing styles of different au-
thors, i.e., the frequencies of use of individual words and pairs of closely
occurring words. Having learned their styles, a neural network system
can then attribute authorship to works that do not come from its training
set. In 1993, Robert Matthews of the University of Aston in Birmingham
and Thomas Merriam, a Shakespearean scholar, created a neural net-
work that could distinguish between the plays of Shakespeare and those
of Marlowe. They trained their network on examples of writings that
were, without dispute, by Shakespeare, and a similar number that were
by Marlowe. During the training process, whenever their network made
a wrong guess as to authorship, parts of the network were adjusted.
24
By
the end of the training period the network could accurately distinguish
between the two playwrights.
Another popular learning technology in the field of AI, genetic al-
gorithms,
25
is also employed in stylometry. In 1995 David Holmes
and Richard Forsyth of the University of Luton in England, applied ge-
netic algorithms to the question of the authorship of the disputed Fed-
eralist papers. They started by creating a set of rules for determining
23
The “disputed papers” have been employed as a testbed for many research projects in stylometry.
24
It was the strengths of the links in the network that were adjusted, see the section “Artificial
Neural Networks” in Chapter 6.
25
See the section “Genetic Algorithms” in Chapter 6.
