Supporting data for "Probabilistic Memory Prioritization Mechanisms in Statistical Learning"

Human learners form conceptual knowledge from statistical learning, an ability of abstracting multiple statistical information across a continuum of probability levels. However, with limited memory and computational resources, how the learning system cope with environmental inputs that embed with multiple forms of information, and multiple types of statistics remains unclear. This thesis investigated this question through three empirical studies. The first study (Chapter 2) developed a novel learning-memory representation paradigm to track the working memory representations of two types of information (i.e., item-specific and abstract) information across high, moderate, and low probabilities during online statistical learning. The second study (Chapter 3) used the electroencephalography approach to further investigate the neural encoding of the abstract and item-specific information across probability levels and non-statistic inputs. The third study (Chapter 4) investigated the online statistical learning of two types of statistics, conditional and distributional. The results show a complex interplay between conditional and distributional learning, regulated by inputs’ probability, structure, and temporal processing. The dataset comprised the dataset for the behavioural experiments in Study 1 (N = 313), the neural data for Study 2 (N = 100) , and the behavioural experiments in Study 3 (N = 245).