Abstract
How do infants see the world? Are they drawn to the same faces and objects that engage adults and older infants, or is attention captured instead by low-level features such as edges, color, and motion? We addressed these questions by recording eye movements of 3-, 6-, 9-, and 12-month-olds infants and adults as they watched a 4-minute video consisting of alternating clips from A Charlie Brown Christmas and moving random-dot kinematograms.
To investigate the focus of attention, we estimated the density of fixations on each frame for each group (using a Bayesian mixture model to cluster fixations) and calculated the entropy of this distribution, measuring the degree to which individuals in each group looked at the same places. While there were no major developmental changes in this measure on the kinematograms, there was a steady increase in consistency of attention across development for the Charlie Brown movies. Put another way, given a complex, social stimulus, adults virtually all look to the same locations, while infants are much more varied in their point of gaze, suggesting a more protracted time-course of development for visual attention than has previously been observed in infant experiments.
We then examined the targets of infants' and adults' eye movements during the Charlie Brown movies using meaningful regions-of-interest (faces, hands) and image-based saliency-maps. In the ROI analysis, we found developmental increases extending past the first year in time spent looking at faces, while in the saliency-map analysis, we found that younger infants' saccades were better predicted by bottom-up image salience than were the saccades of adults. Taken together, these data suggest that diffuse attention driven by bottom-up salience slowly gives way to more focused attention guided by top-down preferences for faces and other meaningful objects over the course of the first year and beyond.
This research supported by a Jacob K. Javits Graduate Fellowship to the first author, NIH R01-HD48733, NIH R01-HD40432, and McDonnell Foundation Grant 412478-G/5-29333.