線形判別式による文体の計量

This paper deals with subjects of authorship attribution through statistical analysis. Critics often assume that there are stylistic differences between authors, but it is obscure, how to discriminate the styles. In this paper, 12 works by two German writers, Theodor Storm and Adalbert Stifter, are analysed through one of the multivariate analysis methods, Linear Discriminant Analysis. The basis for analysis is a word list of 8 high frequent particles in the 12 works, which is made by Java and Perl programs. Based on the list, ten of the works are used to compute discriminants. The rest two works are set aside for validation estimate. And the difference between the styles are computed with Mahalanobis distance. Various equations can be composed according to combinations of selected variables, and they are compared based on statistical tests. The best equation is a non-linear one, but for simplicity, a second best equation, which is linear, is preferred. This equation, using two variables, manages to discriminate the two authors, and a validy of Discriminant Analysis for literary styles is confirmed.

公開日:2010年1月24日で登録したコンテンツは、国立情報学研究所において電子化したものです。