Stylistic Fingerprints, POS-tags, and Inflected Languages: A Case Study in Polish

In stylometric investigations, frequencies of the most frequent words (MFWs) and character n-grams outperform other style-markers, even if their performance varies significantly across languages. inflected languages, word endings play a prominent role, hence different forms cannot be recognized using generic text tokenization. Countless make sparse, making statistical procedures complicated. Presumably, applying one NLP techniques, such as lemmatization and/or parsing, might increase classification. The aim this paper is to examine usefulness grammatical features (as assessed via POS-tag n-grams) lemmatized in recognizing authorial profiles, order address underlying issue degree freedom choice within lexis grammar. Using corpus Polish novels, we performed series supervised authorship attribution benchmarks, compare classification accuracy for types lexical syntactic style-markers. Even POS-tags well was notoriously worse than that markers, difference not substantial never exceeded ca. 15%.