The meaning of salience and the salience of meaning: Object prioritization within naturalistic scenes

In visual scenes, objects can be described in terms of low-level visual salience and higher-level semantic meaning. Understanding how object salience and object meaning influence eye-movement behavior remains a central question in scene perception. Previous work has emphasized one or the other, with recent work suggesting that objects are selected for fixation primarily based on meaning. Building on this work, we fully annotated objects in images of real-world scenes and created meaning maps to estimate object meaning. Because object meaning and visual salience are inherently correlated, we used mixed-effects models to estimate their independent contributions to (a) fixation selection and (b) fixation duration, while accounting for variance at the subject, scene, and object levels. Critically, we examined how results vary across three object annotation methods. Additional multiverse-style analyses considered the choice of saliency model, operationalization of salience and meaning, and eye-movement measures. Data from a scene-memorization task showed significant independent effects of object meaning and object salience. The fixation-selection models yielded a significant negative interaction, suggesting that the attentional advantage of a highly meaningful object is reduced when it is also highly salient, and vice versa. More broadly, the findings show that asking whether meaning “outperforms” salience in predicting fixations oversimplifies the issue. Instead, the results reveal a complex interplay in which their relative influence also depends on specific methodological choices.