The process by which emotion concepts are learned is largely unexplored. Hoemann, Devlin, and Barrett (2020) and Shablack, Stein, and Lindquist (2020) argue that emotion concepts are learned through emotion labels (e.g., “happy”), which cohere variable aspects of emotions into abstract, conceptual categories. While such labeling-dependent learning mechanisms (supervised learning) are plausible, we argue that labeling-independent learning mechanisms (unsupervised learning) are also involved. Specifically, we argue that infants are uniquely situated to learn emotion concepts given their exceptional learning abilities. We provide evidence that children learn from complex, irregular input in other domains (e.g., symbolic numbers) without supervised instruction. Thus, while labels undoubtedly influence emotion concept learning, we must also look beyond language to create a comprehensive theory of emotion concept development.
Multi-Path and Group-Loss-Based Network for Speech Emotion Recognition in Multi-Domain Datasets