An Authorship Attribution for Serbian

An authorship attribution is a problem of identifying the author of an anonymous or disputed text if there is a closed set of candidate authors. Due to the richness of natural languages and numerous ways of expressing individuality in a writing process, this task employs all the sources of language knowledge: lexis, syntax, semantics, orthography, etc. The impressive results of n-gram based algorithms have been presented in many papers for many languages so far. The goal of our research was to test if this group of algorithms works equally well on Serbian and if it is a case, to calculate the optimal values for the parameters appearing in the algorithms. Also, we wanted to test if a syllable based word decomposition, which represents a more human like word decomposition in comparison to n-grams, can be useful in an authorship attribution. Our results confirm good performance of an n-gram based approach (accuracy up to 96%) and show the potential usefulness of a syllable based approach (accuracy from 81% to 89%).