Proceedings of the Annual Meeting of the Cognitive Science Society

A central question in affective computing is how emotions should be represented. Emotion datasets typically adopt either categorical labels (e.g., Ekman categories) or dimensional ratings (e.g., Valence–Arousal–Dominance). Although these frameworks are often treated as comparable descriptions of affect, most work relies on only one representation at a time, limiting cross-dataset comparability and leaving their relationship empirically under-specified. This study explores how VAD ratings map onto categorical emotion judgments, and whether this mapping is consistent across individuals. We fit generalized linear mixed models with participant-level random effects using the MSP-Podcast Corpus to predict categorical labels from dimensional ratings. The results showed a significant effect of Valence, Arousal, and Dominance across several emotion categories (Sadness, Happiness, Anger, Fear, and Disgust) with significant participant-level variability. Overall, our results highlight that categorical and dimensional emotion representations are systematically related but their mapping is shaped by individual differences in affective interpretation. This suggests that affective computing models should explicitly account for rater-dependent variation when using dimensional and categorical emotion representations.