Verbal Identity of a Fictional Character: a Quantitative Study with a Machine Learning Experiment
The paper presents a quantitative research of characters' direct speech patterns in Leo Tolstoy's War and Peace. Tolstoy was known to put a lot of emphasis on the language in which his characters express themselves, and conscious modification of their speech is acknowledged as part of the author's literary technique. In an attempt to measure the scope and intensity of such modification, we extracted speech activity instances from the text of War and Peace, associated them with speakers and identified some distinctive features. We then used these features to train a classifier to recognize the speaker according to the speech. Our hypothesis was that if Tolstoy’s characters actually possessed any unique speech characteristics, the classifier would be able to predict the speaker with some tolerable accuracy.