Stories touch our lives, shape our identities and change how we view the world. They can be factual or fictional, text-based or visual and can take many forms—from novels and non-fiction to conspiracy theories, rumors and disinformation. We can characterize stories by their plot, their characters, their audience, their style, their themes or their purpose. Given the massive power of stories to alter the course of society, innovative methods to understand them empirically are necessary.
In the midst of evaluating submissions for our interdisciplinary Call for Papers titled Science of Stories, which seeks new works that explore the nature of narrative and communication using quantitative and computational methodologies, we reached out to a selection of the Guest Editors for the call–Matthew Jockers, Mirta Galesic, and Mohit Iyyer–to discuss the science behind storytelling.
1. What can we learn from computational approaches to story that we might not learn from other forms of analysis?
Matthew Jockers – A great advantage of computational text analysis is that it facilitates access to different scales: both the micro- and macro-scale. At both of these scales, computers are able to detect and create apparent patterns that a human reader might easily miss. The familiar adage about “not seeing the forest for the trees” is applicable here. Not only does computation allow us to see the trees, it allows us to notice the bark on the trees, and the animals skittering along the branches. It reveals to us the granular details. At the same time, the computer allows us to observe macro-scale patterns; not just the forest, but how the forest is nestled in valleys, clustered near rivers, or how villages around it shelter in its shade– how a story fits into the ecosystem around it.
Mirta Galesic – Computational approaches, when combined with a deep understanding of the story and its cultural and historical context, can tell their own story: a detailed, moment-by-moment analysis of semantic and emotional narratives, their internal dynamics, and their similarities and differences when compared to stories of other authors, cultures, and times.
Mohit Iyyer – Computational story analysis has advantages over manual human analysis in both scale and speed. Computers can process huge amounts of data much faster than a person who has to read and understand every word on the page. While the observations produced by a computer are much more primitive than those of a real domain expert, we can nevertheless use them as a starting point to analyze patterns that are only apparent at a much larger scale (e.g., millions of Tweets, books, blog posts, etc.), which is very exciting!
2. What kinds of research are you most excited about in this area?
Matthew Jockers – I am biased by an interest in plot because it is something that I have worked on in the past.
Mirta Galesic – So many strands of research are incredibly exciting, from analyses of social and semantic networks within stories, to uncovering common ways of thinking and feeling across different corpora, to understanding how stories react to other concurrent and older narratives and societal changes. Work on cognitive and social mechanisms underlying how we process, evaluate, and spread stories is also very important.
Mohit Iyyer – As a researcher in natural language processing (NLP), my primary interest is to advance the language understanding capabilities of machine learning models. I would like to see cutting-edge language representation models adapted to long-form narratives, focused on building robust representations of characters, events, and relationships. As research in understanding stories progresses, I am particularly interested in building computational models that generate stories.
3. Why do you think it is important to examine stories and storytelling in particular? In what ways do you think contributions to this Call for Papers will offer new perspectives on real-world issues or contemporary problems?
Matthew Jockers – This could be a very long answer from a very tall soap box. We, as humans, naturally love a good story. It is why we meet someone and ask them, “What’s your story?” Stories, broadly defined, are everywhere. They are the currency of the realm in every conversation, every negotiation, and every interaction we have as human beings. Stories influence the making of all great decisions, and how a story is told can influence what decisions are made. This is not confined to fiction. I recently read a set of twelve papers for a text-mining conference. Some told the story of their research very well, and those are the papers I remember. Figuring out what makes a good story is one of the most important and essential questions there is. Homer understood that stories were important, and seductive. When Ulysses finally returns from the wars, when he has vanquished all the suitors, what does he do? He sits up in bed with Penelope, telling her stories.
Mirta Galesic – Understanding how we process stories is immensely important in this era of increasing skepticism towards science and other once-established authority figures. Stories are the preferred way in which we communicate information to each other, and today, stories can be spread faster and wider than ever before. People learn from each other’s stories about issues important for their health, society, and our planet; issues from vaccination to immigration and climate change. Scientists need to empower citizens to seek and recognize truthful stories and trustworthy sources, as well as weed out the tales that are seductively believable, but deceiving, and even harmful.
Mohit Iyyer – Stories are an underappreciated domain in the field of natural language processing, which focuses primarily on newswire text. They contain a variety of interesting linguistic phenomena, such as discourse-level narrative arcs and figurative language, which are beyond the capabilities of current state-of-the-art NLP technology. Stories are certainly an interesting domain in themselves, but making progress in understanding them really entails making progress in computational language understanding as a whole.
4. Why do you think reproducibility of work is important to this area of inquiry? What do you think the main barriers are for researchers?
Matthew Jockers – In an emerging area of study, reproducibility is especially important. We are charting new territory, and we are going to make mistakes. Let’s find them and fix them in a spirit of mutual discovery.
Mirta Galesic – Qualitative analyses of stories provide important in-depth insights that cannot be achieved by purely computational approach. At the same time, qualitative analyses are difficult to replicate – researchers’ specific expertise and experience will color their interpretation of the story. Quantitative approaches allow for easy replications of a good number of the studies we conduct on stories, and validation of some predictions of the qualitative studies.
Mohit Iyyer – Reproducibility is important in any research area. Other researchers need to be able to replicate or build upon your work to make progress as a community. Within my field, there is usually no reason not to release material needed to reproduce a particular result (e.g., code or data) as it is usually such a low-cost thing to do. While it takes time to clean up code and write some documentation, and you also need to commit to maintaining the released materials and possibly updating them, doing so really eases the barrier of entry for other researchers who are interested in following up on your work. Another nice feature to add to papers is a “negative results” section, where you can explain what you tried that did not work, so other people do not spend their time on fruitless directions.
5. What advice would you give an early career researcher when preparing to submit to this Call for Papers?
Matthew Jockers – Make it a good story.
Mirta Galesic – There are many ways one can prepare a paper for this Call. One can start from an important societal problem that has to do with stories, and then try to illuminate it and suggest solutions. One can also start with an unresolved theoretical problem and use innovative methodologies to attempt to solve it. In any case, it is useful to provide all the necessary information that other researchers might need when replicating and extending one’s findings.
Mohit Iyyer – Make sure that your paper tells a good story (about stories!) so readers from any field can easily understand what problem you’re tackling and why it is important.
Interested in submitting research that explores the science of stories to PLOS ONE? Take a look at our call for papers announcement. We look forward to seeing your work!
Guest Editors for the Science of Stories Call for Papers
Peter Sheridan Dodds
Guest Editor, PLOS ONE
Peter is a Professor at the University of Vermont’s Department of Mathematics and Statistics. He is Director of the Vermont Complex Systems Center and co-runs the center’s Computational Story Lab. Having a general interest in stories and narratives, complexification, contagion, and robustness, Dodd’s research focuses on system-level, big data problems of all kinds, often networked, sociotechnical ones. His work has been supported by an NSF CAREER award to study sociotechnical phenomena, the McDonnell Foundation, the Office of Naval Research, NASA, the MITRE Corporation, Computer Associates, and Mass Mutual.
Guest Editor, PLOS ONE
Mirta is a Professor and Cowan Chair in Human Social Dynamics at the Santa Fe Institute and Associate Researcher at the Harding Center for Risk Literacy at the Max Planck Institute for Human Development in Berlin, Germany. She studies how simple cognitive mechanisms interact with social and physical environments to produce seemingly complex social phenomena. Her projects focus on developing empirically grounded computational models of social judgments, social learning, collective problem solving, and opinion dynamics.
Guest Editor, PLOS ONE
Mohit is an Assistant Professor in computer science at the University of Massachusetts, Amherst. Previously, he was a Young Investigator at the Allen Institute for Artificial Intelligence. Mohit obtained his PhD at the University of Maryland, College Park, advised by Jordan Boyd-Graber and Hal Daumé III. His research interests lie in natural language processing and machine learning. Much of his work uses deep learning to model language at the discourse level, tackling problems like generating long coherent units of text, answering questions about documents and understanding narratives in fictional text.
Guest Editor, PLOS ONE
Matthew is Dean of the College of Arts & Sciences and Professor of English and Data Analytics at Washington State University in Pullman, WA. Jockers has been leveraging computation to understand narrative and style since the early 1990s. His books on the subject include Macroanalysis: Digital Methods and Literary History, Text Analysis with R for Students of Literature, and The Bestseller Code. In addition to his academic work, Jockers helped launch two text mining startups and worked as Principal Research Scientist and Software Development Engineer in iBooks at Apple.