Author ORCID Identifier
https://orcid.org/0009-0004-9805-8012
Date of Award
Winter 2-2026
Document Type
Thesis (Ph.D.)
Department or Program
Cognitive Neuroscience
First Advisor
Tor D. Wager
Abstract
How do the brain and the mind represent affect? This thesis examines this question across three levels of analysis—individual brain regions, distributed whole-brain patterns, and conceptual structures across languages—using large-scale multi-study and cross-linguistic datasets. The guiding principle is construct validity: testing what our measures and frameworks actually capture across systematic variations in task, context, and population. At the single-region level, a Bayesian mega-analysis of insular cortex across 36 studies reveals domain-general convergence zones in dorsal anterior insula coexisting with domain-selective zones in surrounding regions, organized in a spatial gradient, with distinct neurobiological properties confirmed through coactivation, cytoarchitectonic, and neurotransmitter profiling. At the whole-brain level, evaluation of 29 multivariate predictive models (“brain signatures”) using data from 66 studies reveals that pain signatures show high specificity and generalizability, but aversive and appetitive signatures show bidirectional confusion, with each responding to studies from the other domain that share the stimulus modality of its training data. Reconstructed encoding patterns and data-driven clustering further show that the brain regions underlying these signatures reflect both affective domain and stimulus modality, indicating that whole-brain affective representations are simultaneously shaped by the target construct and sensory context. At the conceptual level, using English and Korean speakers as a case study of cross-linguistic variation, emotion concept representation shows shared global structure alongside language-specific patterns in local organization. However, neither representations derived from large language models nor appraisal features capture these language-specific patterns, aligning more closely with English representation regardless of target language—suggesting that current frameworks do not adequately account for cross-linguistic variation in how emotions are represented. Together, these findings demonstrate that establishing how affective constructs are represented requires systematic testing across variations in task, context, and population, underscoring the importance of construct validation across levels of analysis.
Original Citation
Kwon, M., Wager, T., & Phillips, J. (2022). Representations of emotion concepts: Comparison across pairwise, appraisal feature-based, and word embedding-based similarity spaces. Proceedings of the Annual Meeting of the Cognitive Science Society, 44(44). https://escholarship.org/uc/item/8vj3d366
Kwon, M., Houlihan, S. D., & Phillips, J. (2025). Cross-Cultural Emotion Concept Representation: A Comparison of English, Korean, and Large Language Model Representations. Proceedings of the Annual Meeting of the Cognitive Science Society, 47(0). https://escholarship.org/uc/item/373553tb
Recommended Citation
Kwon, Mijin, "Convergence and Specificity in Neural and Conceptual Representations of Affect: Investigating Affective Representation across Multiple Levels of Analysis" (2026). Dartmouth College Ph.D Dissertations. 459.
https://digitalcommons.dartmouth.edu/dissertations/459
