RBCorr: Response Bias Correction in Language Models

TL;DR

This paper introduces RBCorr, a simple and effective bias correction method for language models that reduces response biases and improves performance across various models and question formats.

Contribution

The paper presents RBCorr, a novel low-cost bias correction strategy that enhances language model evaluation accuracy by eliminating response biases across multiple datasets and models.

Findings

RBCorr effectively reduces response bias in language models.

Bias correction boosts model performance on closed-response benchmarks.

LogProbs-based correction varies with model, dataset, and prompt format.

Abstract

Language models (LMs) are known to be prone to response biases, which present as option preference biases in fixed-response questions. It is therefore imperative to develop low-cost and effective response bias correction methods to improve LM performance and enable more accurate evaluations of model abilities. Here, we propose a simple response bias correction strategy ($\texttt{RBCorr}$) and test it on 12 open-weight language models using yes-no, entailment, and multiple choice questions. We show that response bias is prevalent in LMs pre-correction and that $\texttt{RBCorr}$ effectively eliminates bias and boosts model performance. We also explore the generalizability of bias behavior across models, datasets, and prompt formats, showing that LogProbs-based correction is highly dependent on all three of these aspects. Overall, $\texttt{RBCorr}$ is an easy-to-use method that can boost the performance of smaller LMs and ensure that LM performance on closed-response benchmarks aligns more closely with their true capabilities.