Revisiting Uncertainty Quantification Evaluation in Language Models: Spurious Interactions with Response Length Bias Results
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866918118099517440 |
|---|---|
| author | Santilli, Andrea Golinski, Adam Kirchhof, Michael Danieli, Federico Blaas, Arno Xiong, Miao Zappella, Luca Williamson, Sinead |
| author_facet | Santilli, Andrea Golinski, Adam Kirchhof, Michael Danieli, Federico Blaas, Arno Xiong, Miao Zappella, Luca Williamson, Sinead |
| contents | Uncertainty Quantification (UQ) in Language Models (LMs) is key to improving their safety and reliability. Evaluations often use metrics like AUROC to assess how well UQ methods (e.g., negative sequence probabilities) correlate with task correctness functions (e.g., ROUGE-L). We show that mutual biases--when both UQ methods and correctness functions are biased by the same factors--systematically distort evaluation. First, we formally prove that any mutual bias non-randomly skews AUROC rankings, compromising benchmark integrity. Second, we confirm this happens empirically by testing 7 widely used correctness functions, from lexical-based and embedding-based metrics to LM-as-a-judge approaches, across 4 datasets x 4 models x 8 UQ methods. Our analysis shows that length biases in correctness functions distort UQ assessments by interacting with length biases in UQ methods. We identify LM-as-a-judge methods as the least length-biased, offering a promising path for a fairer UQ evaluation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_13677 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Revisiting Uncertainty Quantification Evaluation in Language Models: Spurious Interactions with Response Length Bias Results Santilli, Andrea Golinski, Adam Kirchhof, Michael Danieli, Federico Blaas, Arno Xiong, Miao Zappella, Luca Williamson, Sinead Computation and Language Artificial Intelligence Machine Learning Uncertainty Quantification (UQ) in Language Models (LMs) is key to improving their safety and reliability. Evaluations often use metrics like AUROC to assess how well UQ methods (e.g., negative sequence probabilities) correlate with task correctness functions (e.g., ROUGE-L). We show that mutual biases--when both UQ methods and correctness functions are biased by the same factors--systematically distort evaluation. First, we formally prove that any mutual bias non-randomly skews AUROC rankings, compromising benchmark integrity. Second, we confirm this happens empirically by testing 7 widely used correctness functions, from lexical-based and embedding-based metrics to LM-as-a-judge approaches, across 4 datasets x 4 models x 8 UQ methods. Our analysis shows that length biases in correctness functions distort UQ assessments by interacting with length biases in UQ methods. We identify LM-as-a-judge methods as the least length-biased, offering a promising path for a fairer UQ evaluation. |
| title | Revisiting Uncertainty Quantification Evaluation in Language Models: Spurious Interactions with Response Length Bias Results |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2504.13677 |