[PS-2.25] Comparing Conjunctive and Transitional Probabilities in Statistical LearningComparing Conjunctive and Transitional Probabilities in Statistical Learning

Mo, D. 1 & Armstrong, B. 1, 2

1 University of Toronto
2 Basque Center on Cognition, Brain and Language

Transitional probability forms the foundation of most statistical learning studies. However, other statistical regularities can be observed across cognitive domains. We investigated the learning of conjunctive probabilities - where multiple elements are considered together to predict another element. In particular, we constructed visual sequences comprising three elements, in which the combination of the first two elements predicted the third element, and in which the informativeness of the two predictive elements was manipulated. We also examined how learning varied as a function of representational overlap (e.g. shared visual features) among the predicted elements across sequences. This representational structure was intended as the simplest possible analogue to how conjunctive probability manifests in domains such as in ambiguous word comprehension. This statistical structure was compared against standard transitional probability in a control experiment. Using a self-paced statistical learning task and offline tests, we observed that adult participants can acquire conjunctive statistical regularities and are sensitive to differences in representational overlap. These results provide an initial proof of concept for the rapid learning of conjunctive probabilities and how acquiring such statistical regularities differ from transitional probability, opening new possibilities for probing what statistical regularities can be learnt in different domains.