Sana Shehabi¹, Srijita Karmakar¹, Miguel P. Eckstein¹; ¹University of California, Santa Barbara

Introduction: Covert and overt attention improve performance in a variety of basic perceptual tasks: detection, contrast and orientation discrimination, letter identification, and scene categorization (Posner, 1980; Carrasco, 2011). Scene understanding is crucial for everyday visual cognition, but we know little about how attention influences the human ability to comprehend scenes. Here, we assess whether covert and overt attention enhance real-world scene comprehension. Methods: In Experiment 1 (covert attention; N=19), observers completed a modified Posner task under enforced fixation. Ninety percent of randomly sampled trials followed a standard 2-location Gabor orientation-identification (± 5 deg) paradigm (80% valid, 20% invalid peripheral precues, Gabor presentation: 50 ms). The remaining 10% were “scene trials,” where Gabors were replaced with real-world images (300 ms) containing individuals performing distinct actions on each side. A response cue (250 ms) indicated which side’s scene observers should describe (50% response cue validity relative to the precue). Semantic similarity index (SSI) between observers’ descriptions and gold-standard descriptions was quantified using cosine similarity computed from text embeddings. Experiment 2 (overt attention; N=10) used the same design, except observers were free to move their eyes. Results: Both experiments showed significant and comparable cueing effects for Gabor trials (p = 0.90, covert vs. overt). Critically, covert attention did not enhance scene comprehension: SSI did not differ (bootstrap p = .060) between valid (0.609 ± 0.014) and invalid (0.579 ± 0.012) cued scenes. In contrast, overt attention resulted in a strong cueing effect on scene comprehension: SSI were higher for valid (0.662 ± 0.012) than invalid (0.438 ± 0.026) cued scenes (bootstrap p < .001). Conclusion: While covert attention produces benefits for basic perceptual tasks, its influence on the comprehension of real-world scenes is small and less than ~1/8th, relative to overt attention.