Multimodal Methods for Spaces and Places in Novels

One of the fundamental pieces of a novel is the setting. Yet quantifying and extracting spatial relationships in novels is difficult due their imprecise nature. My first project uses computer vision techniques to identify 2600 maps that appear in novels, from 32 million page scans. I found that novels with maps are often rich in spatial language within the text and maps are often depicting real (rather than fictional) places. Further, we often think of highly mobile characters making long journeys from one city or country or another, but what if we count the number of separate movements made? Is the house-maid then more mobile than the pirate? For my second project, I finetuned a large language model to perform this annotation task at a rate similar to human experts to find out the qualities of these movements, including: who is moving, where they are moving to/from, how often they move, and how far they move. These projects build a foundation for future multimodal spatial projects that map out character journeys at scale.

Axel Bax is a third-year PhD student in the Information Science department at Cornell University, advised by David Mimno and Matthew Wilkens. He completed his BA in English & Psychology at Johns Hopkins University, as well as his Master’s in Geographic Information Systems. His interest lies in the Digital Humanities, specifically, understanding aspects of literature at scale. He is especially excited about understanding spaces and places within the novel and often uses multimodal methods.