Schedule of Events | Symposia

Sketchpad Series

Decoding Abstract Concepts: A Neuroimaging Study on the Representational Structure of Relational Categories

Poster Session B - Sunday, March 30, 2025, 8:00 – 10:00 am EDT, Back Bay Ballroom/Republic Ballroom

Anthony Dunn¹ (anthony.t.dunn.gr@dartmouth.edu), Katherine Alfred¹, Nick Ichien³, Brianna Aubrey¹, Sophia Baia², Silvia Bunge², David Kraemer¹; ¹Dartmouth College, ²University of California, Berkeley, ³University of Pennsylvania

The ability to infer relations between seemingly disparate concepts is essential for learning and creativity. Revealing how the representational structure of conceptual knowledge in the brain supports the understanding of analogies between concepts is key to explaining how humans derive generalizable knowledge about the world. Despite previous work associating brain regions with the process of analogical reasoning, little is known about how abstract relations themselves are encoded in the brain. To investigate this question, we collected fMRI data from a task in which participants judged whether two words were semantically related. Each word pair belonged to one of three categories describing their abstract relation (whole:part, place:thing, category:exemplar). These relational categories were orthogonal to the task objective, enabling analysis of their representation independent of task demands or individual word semantics. To this end, we are currently analyzing two sources of data: (1) response times to test for priming effects of successive word pairs belonging to the same relational category and (2) neural data for evidence of representational structure corresponding to abstract relations. Specifically, we will use representational similarity analysis (RSA) to determine whether patterns of neural activity elicited by word pairs within the same relational category are more similar than neural activity patterns elicited by word pairs belonging to different relational categories. To ensure that our results are not driven by semantic representations, we will use a word2vec embedding model as a control. Results will address whether abstract relations are represented in the brain beyond the semantic concepts that comprise them.

Topic Area: THINKING: Reasoning