Stanford Lit Lab Trains Neural Network To Identify Suspense in Stories

A group of researchers at the Stanford Literary Lab at Stanford University have developed a virtual reader using a neural network and trained it to identify suspenseful passages in works of literature.

"The big goal of the research is to try and explain why we feel suspense when confronted with certain aesthetic objects, even if we know the outcome of them," said Mark Algee-Hewitt, assistant professor of English at Stanford and leader of the project, in a prepared statement.

Algee-Hewitt and a group of eight graduate students in the English department began by tracking their own emotional responses to reading. One of the various methods they used was to rate paragraphs on a scale of one to ten based on the level of suspense they felt when reading it, according to a report on Stanford's site. They then compared their ratings. While some of the researchers disagreed on the degree of suspense they felt, they did agree on the passages where suspense increased or decreased.

Using digital humanities techniques, they analyzed the passages and discovered specific words or topics that were associated with the element of suspense. More suspenseful passages tended to include words related to the imagination, the senses and movement, and topics related to violence or dramatic weather.

The researchers took the next step of developing a virtual reader using a neural network — a computer program that can be trained to identify new objects on its own — and trained it to identify suspenseful passages based on the set of criteria they had discovered through their analysis. "The neural network achieved 81 percent accuracy in identifying passages it had never seen before as either suspenseful or non-suspenseful," stated a report on Stanford's site.

The researchers hope to publish a paper on the project.

About the Author

Leila Meyer is a technology writer based in British Columbia. She can be reached at [email protected].

Featured