Seminar in Computational Linguistics: The great read: A mixed-methods approach to literary style in Swedish 20th century novel classics

Abstract: We will present results from an ongoing study of literary style in Swedish 20th century prose. The corpus investigated consists of 339 Swedish novels published 1910–1999 that are well-known and can be considered ‘classic’ in some sense, either as written by canonized authors with high literary prestige, or as written by authors with large readerships, or both. Methodologically, we use multiple linguistic measures to track literary style, such as punctuation, sentence and paragraph length, parts-of-speech shares, lexical diversity, and neologisms and cliché phrases. We also conduct a principal component analysis to visualize literary style as a field and how popularity and prestige relates to this. In addition, we discuss methodological problems, since available NLP tools work poorly when applied to more experimental prose fiction for tasks considered ‘trivial’ or ‘solved’ within the NLP community (e.g. sentence segmentation).