Initial exposure to random input decreases subsequent statistical learning: Behavioral and neural evidence.

Statistical learning, the ability to detect regularities between events, can be understood as an emergent property of predictive and memory-based mechanisms that continuously shape how sensory input is encoded and segmented based on prior experience. However, the extent to which statistical learning is influenced by prior context remains relatively underexplored. This work investigated whether initial exposure to a random syllable stream alters the neural and behavioral mechanisms underlying subsequent artificial language learning. In Experiment 1 (random-first group, N = 41; structure-first group; N = 40), adult participants completed a two-alternative forced-choice task after exposure to a structured stream containing trisyllabic "words," either preceded by a random stream or not. Similarly, Experiment 2 (random-first group, N = 20; structure-first group; N = 20) captured listeners' neural responses to syllable and word frequencies during exposure to the artificial language, providing a real-time index of segmentation processes. Results showed better word segmentation for the structure-first group than for the random-first group in both experiments. These findings that early exposure to randomness can interfere with the neural encoding of regularities suggest that statistical learning is not purely stimulus-driven, but dynamically shaped by learning context.