Proceedings of the Annual Meeting of the Cognitive Science Society

Individuals vary in describing emotions using language, utilizing different vocabulary (emodiversity) and levels of context-specificity (granularity). Although both are linked to well-being, they are rarely compared using natural language from daily life. We addressed this gap by analysing ~25,000 Flemish texts from 102 participants collected via a 70-day experience sampling study. Participants provided verbal descriptions of their current experiences and how they felt. We computed emodiversity via Shannon's index and estimated granularity using a novel deep-learning methodology based on emotion word embeddings. Results indicated that momentary valence was positively associated with positive diversity and negative granularity. Furthermore, higher negative granularity predicted fewer stress symptoms and greater mental well-being. These findings suggest that emodiversity and granularity capture distinct facets of emotional construal, variety versus specificity of conceptual differentiation – demonstrating that natural language analysis provides a window into emotion construction processes and their implications for psychological well-being in everyday settings.