LLMs capture explicit lexical emotion labels but not human judgment uncertainty distributions; in-domain fine-tuning and post-hoc calibration reduce the gap by up to 14% while zero-shot models diverge substantially.
Journal of Artificial Intelligence Research, 72:1385– 1470
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
LLMs Capture Emotion Labels, Not Emotion Uncertainty: Distributional Analysis and Calibration of Human-LLM Judgment Gaps
LLMs capture explicit lexical emotion labels but not human judgment uncertainty distributions; in-domain fine-tuning and post-hoc calibration reduce the gap by up to 14% while zero-shot models diverge substantially.