Teaching computers how to understand stories: 5 questions with Wolfgang Yarlott
BLOG: Heidelberg Laureate Forum
Wolfgang Victor Yarlott, a Young Researcher at the Heidelberg Laureate Forum, studies narrative through the lens of computer science. He completed his undergraduate studies at MIT and recently began a PhD program at Florida International University. He sat down with us to talk about the role computers could play in folklore and cultural studies.
How would you explain what you do in 3 sentences?
My work focuses on getting computers to understand stories. Not only understanding how they’re structured—in terms of rising action, climax, falling action, and conclusion—but also inside of the story, what the contents imply about things like culture. I think thus far, there has been a lack of computational approaches to this problem.
Could you give me an example of what a computer might show us about a story?
I don’t know if a computer would necessarily discover something that a human wouldn’t be able to. The main idea is that by using these computation techniques, we’re able to automatically do this. So we’re able to process a much larger amount of stories. You take a huge set of stories—for example, news stories that are specific to a region or a culture. If you can extract cultural information, cultural knowledge, and imbue the system with an ability to make culturally-aware suggestions and decisions, that can have some really strong impacts.
How did you get interested in stories outside of the Western tradition?
I’m from the Crow tribe, a Native American tribe up in Montana, so there’s a strong tradition of stories. I went to a school that was working toward education of Native American students. My mother would talk about coyote—sort of a trickster who’s very clever and tries to take advantage of things. She’d tell me stories about him.
My undergraduate adviser, Patrick Winston, was concerned that his system had only been tested on Shakespearean stories. He asked if I’d be interested in looking at stories specifically from my tribe. I was really excited to take on the project, because I had heard some stories, but this gave me an opportunity to really dig down into the stories. And also to approach them from a new perspective.
How did you enter those kinds of stories into a computer?
The system required stories to be put into a very specific type of English, a simplified ‘toy’ English. One example was the creation story about old man coyote creating the world. In it, he takes a ball of mud, and he breathes on it three times, and he creates the Earth. You have to put it in really simplified terms. Old man coyote, breathes on mud, turns into Earth. It was a rule-based system that looked for example that subverted some common-sense knowledge that you had to give the system. With a lot of folklore, maintaining the structure is maintaining a lot of the important stuff. But I’m no longer working with that system. I’m now working with full-text English stories.
Have you spoken to anyone who studies folklore from a different perspective?
Not yet, but we definitely have plans to interact with a folklorist. We’re in frequent contact with one folklorist, and there’s definitely plans to bring in people outside the field of computer science. Because it’s not just a computer science problem. It’s a problem that touches multiple disciplines.