Probing the Flexibility and Dimensionality of New Semantic Knowledge

Authors: Steve Wang, Jermal Greenwood, Sarah Solomon

Field of Study: Science, Technology, Engineering, and/or Math

Program Affiliation: SPiNS Lab

Faculty Mentors: Sarah Solomon

Easel: 73

Timeslot: Afternoon

Abstract: The flexibility of semantic dimensionality in the human mind is explored through a computational lens. A three-dimensional continuous space was constructed, containing visual “mushroom” stimuli varying along the three dimensions of cap size, stem length, and color. In a concept learning task, participants will learn to successively and flexibly categorize each mushroom as edible or poisonous on the basis of either one, two, or three dimensions. The main goal is to characterize how this mushroom space dynamically expands and contracts as a result of continuous semantic learning. The first crucial step was to establish the validity of the stimulus space and ensure that its representational geometry—as assessed through similarity ratings—approximates a three-dimensional cube before any semantic learning occurs. Data demonstrating the validity of this stimulus space is provided, revealing its ability to reflect the predicted category bias, expansion, and compression in the planned category learning experiments.