Gaze transitions during encoding predict episodic recall of scenes

A new study by Hugo Rydel and Alex Kafkas, published on arXiv (2606.20844), investigates whether the way our eyes move between objects in a scene can predict how well we remember it. The researchers used scene graph annotations—a technique that captures relationships between objects—to quantify 'relational gaze transitions' from eye-tracking data. Participants viewed complex, naturalistic scenes while their eye movements were recorded, and later attempted to recall details from blank screens. The key finding: during both initial encoding and blank-screen retrieval, participants exhibited above-chance gaze transitions between meaningfully related objects. This indicates that gaze actively tracks the structure of a scene, both when seeing it for the first time and when reconstructing it from memory.

Crucially, the amount of relational scanning during encoding (but not during retrieval) predicted how well participants later remembered both individual objects and their relationships. This effect held even after accounting for visual salience, how often participants fixated on objects, the objects' semantic meaning, and differences between images. The results suggest that relational gaze during encoding serves as a functional marker of how the brain organizes visual information into coherent episodic memories. The study moves beyond simplified displays and isolated relation types, offering a method to measure cognitive organization in naturalistic settings. Potential applications range from diagnostic tools for memory impairments to novel learning and training systems that optimize gaze patterns for better recall.