Submission 710
Harnessing Semantic Networks for Efficient Task Learning
SymposiumTalk-03
Presented by: Senne Braem
Humans are remarkably efficient at learning new tasks, in large part by relying on the integration of previously learned knowledge. However, most of our research on cognitive control typically happens in experiments using abstract and non-tangible stimuli that do not rely on participants' existing semantic knowledge. Here, I will present a series of experiments were we demonstrate how existing, semantically rich distinctions allow for a more robust learning of novel task information. Specifically, through both behavioral analyses and fitting neural network models, we show how pre-existing semantic structures might particularly help with creating more separated task representations, that are more resistant to catastrophic forgetting and can be repurposed to other tasks. Next, I will show how through the use of spatial arrangement tasks, we can tap into pre-existing semantic structures and show how individual differences in sensitivity to semantic dimensions can predict the learning and embedding of entirely new tasks. Finally, I will demonstrate how we can leverage this paradigm to test contemporary theories on the development of cognitive control in prepubescent children. We believe our work helps towards a more integrative understanding of how cognitive control functions are learned and develop through interactions with semantic cognition during novel task learning.