Santolin, C. & Regolin, L. .
Department of General Psychology, University of Padova, Italy
Statistical learning is the ability to track probabilistic structures from the sensory input, in order to organize and interpret the environment. For instance, human infants are capable of extracting statistical information from both linguistic (e.g. artificial languages) and nonlinguistic inputs (e.g. streams of shapes). Besides being robust enough to operate across domains and modalities, statistical learning has also been reported in some nonhuman species, reinforcing the idea of a domain-general learning process. In the present study, we exposed visually-naïve newborn chicks to a visual computer-presented stream of shapes whose ordering was defined by transitional probabilities within/between shape-pairs. No reward has been provided to the animals, enabling us to investigate statistical learning as a form of unsupervised learning. Afterward, we tested chicks' discrimination of the familiar sequence from a random presentation (Exp.1) or from a novel combination (Exp.2) of the familiar shapes. In both experiments, chicks recognized their familiar stimulus suggesting that this species presents an early sensitivity to the probabilistic structure underlying complex visual stimuli. Our results provide the first evidence of visual statistical learning in an avian species, highlighting similar abilities with respect to human infants and promoting the idea of statistical learning as the result of convergent evolution.