Flexible statistical learning across modalities: Online and offline measures reveal different aspects of adaptation to changing regularities

The ability to discern the statistical regularities in our environments has been shown to support key cognitive functions, including attention, prediction and language learning. While most research has focused on stable regularities, real-world patterns often change over time, requiring adaptation. In the context of embedded pattern learning, where continuous input consists of hidden pairs or triplets, prior work showed that learning of an initial structure can hinder learning of an updated one. Alongside post-exposure (offline) learning measures, the current study incorporated online target detection during exposure to gauge real-time learning and adaptation to novel patterns more directly. In three separate blocks, participants were exposed to a stream of embedded pairs that were reshuffled into new pairs halfway through the stream. We administered the same task in both the visual and auditory modality, allowing us to explore modality differences. In the auditory modality, the online target detection measure revealed that participants learned both the initial and updated regularities, but with an advantage for learning the former. In contrast, the offline measure only evidenced recognition of the initial patterns, echoing previously reported primacy effects. In the visual modality, learning was not observed online but was revealed in sensitivity to both sets of regularities in the offline test. Our results provide evidence for flexible statistical learning of different types of sensory regularities under incidental learning conditions, which is important for the challenge of learning in dynamic environments. They also underscore the non-overlapping information that is provided by on- and offline measures.