Zeitschrift für digitale Geisteswissenschaften (Dec 2017)
»Delta« in der stilometrischen Autorschaftsattribution
Abstract
In this article, we present current stylometric studies on Delta. (1) We discuss why the use of cosine similarity improves the rate of success; our experiments on vector normalization lead to a better understanding of how Delta works. (2) Based on a corpus of Middle High German texts, we show that metrical properties can also be used for authorship attribution. The degree to which Delta is influenced by non-normalized medieval spellings is also investigated. (3) Using a corpus of Arabic-Latin translations, we explore how selective feature elimination can be used to separate the translator signal from the genre signal.
Keywords