The target similarity conundrum in rapid serial visual presentation

Abstract

To deal with rapidly changing visual information, the brain has developed an attentional system that optimises our perception to detect important changes. To probe this attentional system, researchers often use rapid serial visual presentation (RSVP) paradigms. In a two-target RSVP, participants are asked to report two targets (T1/T2) that are embedded among multiple distractors in a stream of stimuli. In a well-known RSVP phenomenon, known as repetition blindness (RB), T2 is often unreported when targets are repeated. This effect has been argued to be due to a memory, rather than a perceptual, failure. In a previous study, we measured multivariate brain representations throughout the visual hierarchy and showed that the more similar targets were in high-level regions, the more likely participants were to miss T2, extending previous findings of RB. In contrast, when targets were similar in low-level regions (e.g. V1), performance on T2 increased. In the current study, we aim to resolve this conundrum and hypothesise that V1-similarity between targets increases the rate of evidence accumulation for T2. Based on a large high-resolution functional magnetic resonance imaging dataset (Natural Scenes Dataset, http://naturalscenesdataset.org) we constructed RSVP trials with low, mid, and high V1-similarity between targets. We then asked participants to memorise T1 while making a speeded animacy judgment on T2. We employed drift-diffusion modelling on the reaction time distribution for T2 responses to show that evidence accumulation is positively modulated by V1 similarity. This effect was present even after filtering away trials where T1 was an animal, meaning it cannot be explained by any selection bias of images in either category. In contrast to previous RB theories, associated with repetitions in the task-relevant domain, V1 similarity not only increases T2 performance but appears to be related to pre-conscious processes.