Affective Profiles of Figurative Language: Cognitive and Computational Insights

Abstract Figurative language is a powerful means of expressing complex emotions and thoughts. This study examines how different types of figurative language relate to emotion by analyzing metaphors, similes, idioms, and sarcasm with four BERT-based emotion classification models that vary in emotional granularity. Using two human-annotated datasets, we compare affective patterns across emotion representations ranging from 3 to 28 dimensions. The findings reveal both regularity and variation across figurative types. Sarcasm shows a stable association with negative emotion in both datasets, while the emotional profiles of metaphors, idioms, and similes vary across corpora and annotation frameworks. These findings indicate that some figurative types have more stable emotional signatures than others, though generalizability beyond the examined datasets requires further investigation. By systematically mapping emotional patterns across figurative types, this study contributes to ongoing work in computational linguistics and cognitive-affective research by clarifying how emotion is distributed across figurative forms and by identifying challenges for the development of more emotionally aware language technologies.