Statistical Learning in Language Acquisition

Statistical learning is the ability of humans and other animals to extract statistical regularities from their surroundings in order to learn about their environment, an ability now understood as a generalized learning mechanism even though it was first identified in the context of human infant language acquisition. The earliest evidence for it came from a study by Jenny Saffran, Richard Aslin and Elissa Newport, in which eight-month-old infants were played nonsense streams of monotone speech built from four repeating three-syllable pseudowords; after only two minutes of exposure, the infants reacted differently to the pseudowords than to nonwords built from the same syllables in a different order, showing that they had picked up the statistical relationships between syllables that always occurred together as opposed to syllables that only occurred together rarely. This mechanism is thought to be one way that children work out which groups of syllables form individual words, and since its initial discovery in word learning, the same statistical-learning mechanism has also been proposed as a contributor to phonological and syntactic acquisition, as well as to non-linguistic learning in both primates and non-primates, across both visual and auditory information.