Discriminating the registers and styles in the modern greek language-part 2: Extending the feature vector to optimize author discrimination

George Tambouratzis, Nikolaos Hairetakis, Marina Vasileiou

Abstract: This article describes a method for discriminating among authors within a given register of Modern Greek. The focus here is to determine to what extent the stylistic differences among authors can be detected with a high degree of accuracy for a set of texts belonging to a well‐defined register. To that end, the chosen register is characterized by a well‐defined sub‐language, from which a corpus of more than 1,000 documents has been created. To discriminate the texts according to author style, a series of experiments have been performed using statistical techniques. Each text has been represented by a vector covering several linguistic aspects, in an effort to determine the most effective style markers. The experimental results indicate that the proposed approach can successfully separate the author styles for a given register. An extensive study of the effectiveness of the different variable categories has been performed. For instance, diglossia information on its own is not sufficient for author discrimination. Instead, a systematic evaluation process indicates that part‐of‐speech, structural and algorithmically derived lemma‐frequency variables are the most important style markers, their use leading to an author discrimination accuracy exceeding 90%.