Abstract
Our ability to categorize faces develops rapidly within infancy. Recent behavioral findings suggest a qualitative developmental change: young infants categorize face races based on perceptual similarities (e.g., White vs. Black faces), but older infants’ categorization is driven by conceptual knowledge (e.g., own- vs. other-race faces). The current study examined this developmental change through infants’ eye movement patterns. Eye movement dynamics have been regarded as a reliable measure to index preverbal infants’ perceptual capacities. The current study used a data-driven approach to model the temporal and spatial pattern of infants’ face-scanning data with the Hidden Markov Model (HMM). Infants aged 3 (n = 23), 6 (n = 32), and 9 months (n = 26) watched videos of own-race and other-race female actors (30s/video). With eye movement data, we built one HMM for each participants’ own-race face trial, and one HMM for each participants’ other-race face trial. Then, we used leave-one-subject-out cross-validation to examine the accuracy of face-race classification. If infants formed categories of face-race, the face-race classification should be significantly above chance (50%). The results showed successful classification in 6- (65.5%, p = .002) and 9-month-old infants (72.9%, p = .001), but not in 3-month-olds (47.8%, p = .719). To probe the development of face-race categorization, we performed cross-age classification. The HMMs of 6-month-olds failed to classify face-race from the eye movement of 3- (50.0%, p = .998) or 9-month-olds (50.0%, p = .998). The HMMs of 9-month-olds could classify the eye movement of 6-month-olds (61.9%, p = .023) but not 3-month-olds (58.7%, p = .230). Together, these results show the development of face-race categorization as an incremental process: infants’ face-race categories become increasingly distinct with age. This suggests that young infants’ face race representations serve as building blocks for later refined categories of face-races.