A quantitative study of non-linearity in storytelling

In this paper, we present a study of non-linearity in storytelling in a collection of 2,348 books published since 2001 that are divided among 10 different categories. We employ word embeddings to capture the semantic non-linearity of a book, along with three associated measures called speed, volume, and distance. We find that narrative non-linearity is strongly associated with the communication of non-instrumental (imaginary) information, but only very weakly negatively associated with a book's success. When we decompose non-linearity, we observe that reader preferences for books favor greater narrative distance with lower speed, i.e. books that take readers further but do so more parsimoniously are more successful. Our work aims to contribute to a growing body of studies that use computational methods to better understand human narrative behavior.