Distributional Information Shapes Infants' Categorization of Objects

While distributional learning has been successfully demonstrated for auditory categorization, this study tests whether this mechanism also applies to object categorization: Ten-month-olds (n = 38) were familiarized with either a unimodal or bimodal distribution of a visual continuum. Using automatic eye tracking, we assessed categorization through the alternating/nonalternating paradigm. For infants in the bimodal condition, their average dwell time was larger for alternating trials than for nonalternating trials, while infants in the unimodal condition initially looked equally long at both types of trials. This group difference suggests that the shape of frequency distribution bears on the number of categories that infants construct from a continuum. Later in test, all infants show this alternating preference. We conclude that categorization is a flexible process, continuously adjusting itself to additional input.