%0 Conference Proceedings %T Stop Measuring Calibration When Humans Disagree %A Baan, Joris %A Aziz, Wilker %A Plank, Barbara %A Fernandez, Raquel %Y Goldberg, Yoav %Y Kozareva, Zornitsa %Y Zhang, Yue %S Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing %D 2022 %8 December %I Association for Computational Linguistics %C Abu Dhabi, United Arab Emirates %F baan-etal-2022-stop %X Calibration is a popular framework to evaluate whether a classifier knows when it does not know - i.e., its predictive probabilities are a good indication of how likely a prediction is to be correct. Correctness is commonly estimated against the human majority class. Recently, calibration to human majority has been measured on tasks where humans inherently disagree about which class applies. We show that measuring calibration to human majority given inherent disagreements is theoretically problematic, demonstrate this empirically on the ChaosNLI dataset, and derive several instance-level measures of calibration that capture key statistical properties of human judgements - including class frequency, ranking and entropy. %R 10.18653/v1/2022.emnlp-main.124 %U https://aclanthology.org/2022.emnlp-main.124 %U https://doi.org/10.18653/v1/2022.emnlp-main.124 %P 1892-1915