RAZOR: Sharpening Knowledge by Cutting Bias with Unsupervised Text Rewriting

TL;DR

RAZOR is an unsupervised text rewriting method that reduces dataset biases in language models by iteratively replacing biased segments, leading to improved generalization and fairness without prior bias knowledge.

Contribution

Introduces RAZOR, a novel unsupervised text rewriting approach for debiasing language models, eliminating the need for prior bias information and enhancing model robustness.

Findings

RAZOR improves F1 scores by 3.5% on FEVER and 6.5% on MNLI and SNLI datasets.

Effectively mitigates known biases, halving bias-related terms without prior bias knowledge.

Achieves performance comparable to state-of-the-art models that rely on prior bias information.

Abstract

Despite the widespread use of LLMs due to their superior performance in various tasks, their high computational costs often lead potential users to opt for the pretraining-finetuning pipeline. However, biases prevalent in manually constructed datasets can introduce spurious correlations between tokens and labels, creating so-called shortcuts and hindering the generalizability of fine-tuned models. Existing debiasing methods often rely on prior knowledge of specific dataset biases, which is challenging to acquire a priori. We propose RAZOR (Rewriting And Zero-bias Optimization Refinement), a novel, unsupervised, and data-focused debiasing approach based on text rewriting for shortcut mitigation. RAZOR leverages LLMs to iteratively rewrite potentially biased text segments by replacing them with heuristically selected alternatives in a shortcut space defined by token statistics and positional information. This process aims to align surface-level text features more closely with diverse label distributions, thereby promoting the learning of genuine linguistic patterns. Compared with unsupervised SoTA models, RAZOR improves by 3.5% on the FEVER and 6.5% on MNLI and SNLI datasets according to the F1 score. Additionally, RAZOR effectively mitigates specific known biases, reducing bias-related terms by x2 without requiring prior bias information, a result that is on par with SoTA models that leverage prior information. Our work prioritizes data manipulation over architectural modifications, emphasizing the pivotal role of data quality in enhancing model performance and fairness. This research contributes to developing more robust evaluation benchmarks for debiasing methods by incorporating metrics for bias reduction and overall model efficacy.