Improving Semantic Uncertainty Quantification in Language Model Question-Answering via Token-Level Temperature Scaling
This addresses the need for reliable uncertainty quantification in language models for question-answering tasks, but it is incremental as it builds on existing temperature scaling techniques.
The paper tackled the problem of poor calibration in semantic uncertainty quantification for language model question-answering by showing that current fixed-temperature heuristics produce miscalibrated and poorly discriminative confidence distributions, and demonstrated that optimizing a single scalar temperature consistently improves calibration, discrimination, and downstream entropy, outperforming baselines and more expressive methods.
Calibration is central to reliable semantic uncertainty quantification, yet prior work has largely focused on discrimination, neglecting calibration. As calibration and discrimination capture distinct aspects of uncertainty, focusing on discrimination alone yields an incomplete picture. We address this gap by systematically evaluating both aspects across a broad set of confidence measures. We show that current approaches, particularly fixed-temperature heuristics, produce systematically miscalibrated and poorly discriminative semantic confidence distributions. We demonstrate that optimising a single scalar temperature, which, we argue, provides a suitable inductive bias, is a surprisingly simple yet effective solution. Our exhaustive evaluation confirms that temperature scaling consistently improves semantic calibration, discrimination, and downstream entropy, outperforming both heuristic baselines and more expressive token-level recalibration methods on question-answering tasks.