Bridging Real-world Summarization and Laboratory-based Memory Recall

A fundamental question in psychology is how humans digest, retain, and reconstruct information. While this process has been extensively studied in laboratory-based memory recall tasks, we have missed opportunities to examine it under more naturalistic, real-world conditions. Summarization is a task that people routinely engage in every day, and it is also a long-standing challenge in the machine learning community. Researchers in machine learning have developed rich datasets of human summaries for model training, which represent a valuable and largely untapped resource for cognitive science researchers. In this work, we analyze 45,220 human summaries drawn from three large datasets, including news articles, autobiographical stories, and Reddit posts, and compare them with a dataset of narrative memory recall. We demonstrate a close relationship between human summarization and laboratory-based memory recall. We show that while summarization operates at a higher level of the event abstraction than memory recall, the two tasks share deep similarities. The process of summarizing is fundamentally constrained by the mechanisms of memory recall, displaying behavioral patterns that align with well-known memory biases. Conversely, memory recall takes the form of summarization, prioritizing more central and important information. Together, these findings bridge the real-world task of summarization and the rich literature on memory recall and lay the foundation for developing integrated theories of how humans digest and reconstruct information.