Towards Robust NLG Bias Evaluation with Syntactically-diverse Prompts

TL;DR

This paper proposes a method for evaluating biases in NLG systems using syntactically-diverse prompts, which leads to more reliable and tone-invariant bias assessments compared to fixed templates.

Contribution

It introduces a paraphrasing approach to generate syntactically-diverse prompts for bias evaluation, improving robustness over traditional fixed-template methods.

Findings

Syntactic variation affects bias measurement outcomes.

Some structures induce more toxic content, others less biased.

Robust evaluation benefits from tone-invariant, diverse prompts.

Abstract

We present a robust methodology for evaluating biases in natural language generation(NLG) systems. Previous works use fixed hand-crafted prefix templates with mentions of various demographic groups to prompt models to generate continuations for bias analysis. These fixed prefix templates could themselves be specific in terms of styles or linguistic structures, which may lead to unreliable fairness conclusions that are not representative of the general trends from tone varying prompts. To study this problem, we paraphrase the prompts with different syntactic structures and use these to evaluate demographic bias in NLG systems. Our results suggest similar overall bias trends but some syntactic structures lead to contradictory conclusions compared to past works. We show that our methodology is more robust and that some syntactic structures prompt more toxic content while others could prompt less biased generation. This suggests the importance of not relying on a fixed syntactic structure and using tone-invariant prompts. Introducing syntactically-diverse prompts can achieve more robust NLG (bias) evaluation.