ABSTRACT
Scholars in the humanities have long argued over the authorship of works ranging from Elizabethan histories to religious texts. Most of the debate has been essentially subjective and qualitative. However, the advent of computer technology has led to the development of stylometry—the quantitative analysis of literary style based on, for example, frequency of words used by different authors. With their ability to cope with both nonlinear and noisy data sets, neural networks are well suited to the stylometric problem. Here we show that they out-perform linear methods of identifying authors, and illustrate their power with studies of disputed works from the era of William Shakespeare.
