Analysis of Stylometric Variables in Long and Short Texts

This paper presents some experiments in the task of authorship attribution. We achieve this task by a stylometric analysis of some stylistic markers tested in two Spanish corpora. The first corpus is composed of long texts written by professional authors, while the second corpus is formed by short texts written by students. In both corpora, different text genres are included. Thus, the objective of this study is to analyze several stylometric variables to test its capacity as markers for authorship attribution when the corpora vary in size and text genre. We represent the texts as high dimensional vectors and we visualize the similarities between them using multidimensional scaling. We conclude that the length of texts is a factor that affects the discriminatory capacity of the stylometric variables. We also found that there are certain variables that are better than others to identify specific authors and specific text genres.