On August 19th, researchers from Disney and the University of Massachusetts Boston published the results of a fascinating project merging the disciplines of science and literature. Researchers used AI to glean the quality of short narratives by training neural networks to analyze tens of thousands of story-like posts from Quora.com and predict which of the site’s stories stood out most based on upvotes. More from ScienceAlert below…
Can they crack the magic of storytelling with a neural network?
“Our neural networks had some success in predicting the popularity of stories,” says one of the team, Boyang Albert Li from Disney Research. “You can’t yet use them to pick out winners for your local writing competition, but they can be used to guide future research.”
Three neural networks were then developed: one to look at individual chunks of the stories, one to analyse how those chunks fit together, and one to assess the stories more generally. As a combination, the networks were able to learn what separated the better stories from the rest.
The test was to see if the neural networks, once trained, could take a new story from the database and predict how many Quora upvotes it got – a sign that the AI was understanding what makes a story popular and what doesn’t.
“The ability to predict narrative quality impacts on both story creation and story understanding,” says Disney Research vice president Markus Gross, who wasn’t directly involved in the research.
“To evaluate quality, the AI needs some level of understanding of the text. And if AIs are to create narratives, they need to be able to judge the quality of what they are producing.”
Photo credit: Disney