A Computational Approach for Authorship Attribution on Multiple Languages
Paulo Junior Varela, Michel Albonico, Edson Jose Rodrigues Justino, Flávio Bortolozzi · 2018
In this paper, we describe a computational approach for authorship attribution of literary texts in a multilingual environment. First, we define a set of syntactical features at text structural levels by considering the internal grammatical structure of sentences. The main idea is to extract syntactic functions from every word necessary to make up a sentence, such as subject and predicate. Such elements denote a writing pattern, tracking a stylometric profile for each author. Second, we gather literary texts in five languages: Portuguese, Spanish, French, German, and English. We analyze distinct scenarios through a series of experiments, taking into account writer-dependent and writer-independent approaches while using authorship verifica- tion and identification strategies. Furthermore, we evaluate the model behavior when the amount of information in each sample is varied and the impact of the reference samples. We use a support vector machine classifier in this process.