Statistical learning of spatiotemporal regularities dynamically guides visual attention across space

Delayed ocular disengagement from arousing scenes

Visual exploration of the world is supported by eye movements which can be speeded up or delayed depending on bottom-up stimulation, top-down goals, and prior associations. Previous studies observed faster initiation of saccades toward emotional than neutral natural scenes; however, less is known concerning saccades which originate from emotional, compared with neutral, scenes. Here, we addressed this issue by examining a task in which participants continuously moved their gaze from and toward pictures (natural scenes), which could be emotional or neutral, and changed position in every trial. Saccades were initiated later when the starting picture was emotional compared to neutral, and this slowing was associated with the arousal value of the picture, suggesting that ocular disengagement does not vary with stimulus valence but is affected by engaging picture contents such as erotica and threat/injuries.

Attentional suppression is in place before display onset

Recent studies have shown that observers can learn to suppress a location that is most likely to contain a distractor. The current study investigates whether the statistically learned suppression is already in place, before, or implemented exactly at the moment participants expect the display to appear. Participants performed a visual search task in which a distractor was presented more frequently at the high-probability location (HPL) in a search display. Occasionally, the search display was replaced by a probe display in which participants needed to detect a probe offset. The temporal relationship between the probe display and the search display was manipulated by varying the stimulus onset asynchronies (SOAs) in the probe task. In this way, the attentional distribution in space was probed before, exactly at, or after the moment when the search display was expected to be presented. The results showed a statistically learned suppression at the HPL, as evidenced by faster and more accurate search when a distractor was presented at this location. Crucially, irrespective of the SOA, probe detection was always slower at the HPL than at the low-probability locations, indicating that the spatial suppression induced by statistical learning is proactively implemented not just at the moment the display is expected, but prior to display onset. We conclude that statistical learning affects the weights within the priority map relatively early in time, well before the availability of the search display.