Fig. 7: Forced fixation at SU-relevant and SU-irrelevant locations.

a A between-subjects (N=15) experimental design was used for the forced fixation experiment. Participants were instructed to maintain fixation on SU-relevant or SU-irrelevant objects for 500 ms, while real-time eye-tracking ensured fixation was maintained. They were asked to provide a best-guess description of what was happening in the scene. b Examples show that observers' descriptions were more similar to the gold standard meaning of a scene when they fixated on the SU-relevant objects. c The scatter plot shows the similarity scores (Gemini, see Figure Supplementary 1b, c for comparison of Gemini similarity scores and human ratings) of the scene descriptions relative to the gold standard for each image (averaged across observers) while fixating SU-relevant objects vs. while fixating SU-irrelevant objects. d Average similarity score relative to the gold standard description for the descriptions provided by participants fixating on either SU-relevant or SU-irrelevant (15 participants per fixation location per image, 36 images). The similarity of the descriptions was significantly higher (p < 0.001) when participants were fixating at the SU-relevant locations compared to the SU-irrelevant locations. The top dashed line indicates an upper bound established by the average similarity score across gold standards (inter-observer agreement of descriptions with unlimited time). The bottom dashed line indicates a lower bound calculated by permuting descriptions across images and comparing them to the gold standard from unmatched images. ***** (p < 0.00001). A one-tailed bootstrapped analysis was conducted to test the significance of all the results. In d, the line within the box indicates the median. The box spans the interquartile range (IQR), and the whiskers extend to the most extreme values within 1.5 × IQR across images. Consent was obtained from the person in the featured image for its publication.