“Attentional fingerprints” in conceptual space: Reliable, individuating patterns of visual attention revealed using natural language modeling

The eyes are a window into the mind. Eye-tracking studies in the psychology literature report large individual differences in how people deploy attention when scanning photographs of real-world environments. Yet, a key question remains unanswered. Are the eyes a window into a specific mind? In other words, are individual differences in gaze allocation unique to a person and stable within that person? If so, what representational space structures these individual “attentional fingerprints”? Here, we developed a novel approach for describing abstract semantic information present in real-world scenes using a computational language model. We then measured participants’ (N = 42) naturalistic attention while they actively explored real world environments in head-mounted virtual reality. For each participant, we modeled the relationship between their attentional patterns and an abstract semantic feature space (using a natural language processing model, BERT) in N-1 scenes, and iteratively predicted attention in a left-out scene. In brief, we find evidence for reliable, individuating patterns of attention (i.e., “attentional fingerprints”) in abstract semantic space.