Despite being unbound by space and time, fictional protagonists in American literature travel fewer miles than their nonfiction counterparts, according to a Cornell-led research team that used artificial intelligence to analyze nearly 13,500 books from the last 230 years.

The team also found that despite social constraints that affected women’s mobility over the last three centuries, female protagonists throughout American literature travel just as much as male protagonists.

The findings highlight novel ways in which scholars in the digital humanities and cultural analytics use large language models – the AI-powered tools behind technology such as ChatGPT – on vast libraries of digitized books to unearth granular details about culture and life throughout history.

“In aggregate, if you understand what’s happening in literature, you also understand something about what’s happening in the world,” said Matthew Wilkens, associate professor of information science in the Cornell Ann S. Bowers College of Computing and Information Science. “You would imagine that the mandate of fiction is to explore more broadly than what actual daily life allows. It turns out that’s not what fiction does. Fictional worlds are smaller than nonfictional worlds.”

Wilkens is the lead author of “Small Worlds: Measuring the Mobility of Characters in English-Language Fiction,” published Sept. 26 in the Journal of Computation Literary Studies (JCLS) and presented in June at JCLS’s Conference of Computational Literary Studies.

The researchers introduced the AI-powered model they developed to analyze a digital corpus containing 13,383 American fiction and nonfiction books published between 1789 and 2021. The model is savvy enough to identify characters and accurately measure when, where and how far they physically traveled.

They found that real-life protagonists in nonfiction literature such as histories and biographies cover twice as much geographic distance as do fictional protagonists, which surprised Wilkens. Traveling can be a hassle in real life – especially in the 18th and 19th centuries, when safe and reliable transportation wasn’t as accessible – while fictional characters can move about at will with a dash of a writer’s pen or keystroke.

The team also found that compared to fiction, nonfiction narratives included more “sites of power” – ranging from the White House, government offices and “palaces” to workplaces, prisons and warzones. Fictional narratives were more often set in domestic and semi-public spaces like apartments, kitchens and bars, the researchers found.

Mining literature penned throughout the centuries offers researchers a lens into everyday life, with richer detail than historical news accounts, Wilkens said.

“This work is a way to get a view into the more common or daily experience that just isn’t well attested in a lot of other historical documents,” Wilkens said. “You’re not going to survey millions of people in 1950 or 1850 about what their life was like, and news accounts don’t say much about ordinary people. One of the big traces we have are the stories they left behind.”

Paper co-authors are Elizabeth F. Evans, associate professor of English at Wayne State University; Sandeep Soni, assistant professor in the Department of Quantitative Theory and Methods at Emory University; David Bamman, associate professor in the School of Information at the University of California, Berkeley; and Andrew Piper, professor in the Department of Languages, Literatures, and Cultures at McGill University.

This research was partly supported by the National Science Foundation and the National Endowment for the Humanities.

Louis DiPietro is a writer for the Cornell Ann S. Bowers College of Computing and Information Science.