A team from the Institute for Advanced Study, Emory University in the US, and the Weizmann Institute of Science, Israel, has developed a new mathematical framework to understand how humans store meaningful narratives in memory.
Their approach uses random trees, mathematical objects that can represent branching structures, to model the way people remember stories.
The study’s lead authors point out that their the goal was to create a rigorous theory of human memory for complex material stories. Published in the journal Physical Review Letters, the research combines concepts from mathematics, computer science, and physics to explore how events and details in narratives are connected in the mind.
Researchers find predictable patterns in how we recall stories
While many scientists believe narratives are too complex for a unified mathematical theory, the latest study shows otherwise. According to senior author Misha Tsodyks, despite the complexity of stories, there are statistical patterns in how people recall them, which can be predicted using a few simple underlying principles.
“We introduce a statistical ensemble of random trees to represent narratives as hierarchies of key points, where each node is a compressed representation of its descendant leaves, which are the original narrative segments,” the researchers note in the abstract.
Tsodyks and his team tested their random tree memory model by conducting online recall experiments with over 100 participants recruited through Amazon and Prolific. The team used 11 narratives of different lengths, from 20 to 200 sentences, originally compiled by American linguist William Labov. Participants were asked to recall these stories, and the researchers analyzed their responses to see if their theory held true.
Following this, the team used spoken narratives recorded by Labov in the 1960s and analyzed the large amount of data by relying on modern tools like AI and large language models.
The researchers found that people often summarize entire episodes of a story into single sentences, leading to the conclusion that narratives are stored in memory as tree structures. In this model, nodes closer to the root represent broader summaries, while more detailed events branch out further away.
Brain stores narratives as mathematically predictable trees
The study authors believe that when someone first hears or reads a story and understands it, their brain constructs a tree-like structure to represent the narrative. Since people interpret stories differently, each person’s memory tree has a unique structure.
To test the idea, Tsodyks and his team created a model based on ensembles of random trees with specific structures. What they found was that this model could be solved mathematically and its predictions matched experimental data, with the main insight being that all meaningful material, like narratives, may be represented in memory in a similar tree-like way.
The researchers now believe their findings have broader implications for understanding human cognition, since narratives are a common way people make sense of their personal experiences as well as social and historical events.
Furthermore, such discoveries also highlight the potential of combining mathematical models with AI techniques to better study how meaningful information is stored and organized in memory.
You can view the study here.