通过PRM引导的候选选择和顺序细化缓解英语和乌尔都语语言模型中的社会偏见
Mitigating Social Bias in English and Urdu Language Models Using PRM-Guided Candidate Selection and Sequential Refinement
AI总结：
本文通过PRM引导的候选选择和顺序细化方法，缓解英语和乌尔都语语言模型中的社会偏见，展示了在不同语言中偏见减少和公平性提升的效果。
摘要：
大型语言模型（LLMs）越来越多地调解人类交流、决策支持、内容创作和信息检索。尽管流利度令人印象深刻，这些系统经常产生有偏见或刻板印象的内容，尤其是在被社会敏感语言提示时。越来越多的研究表明，此类偏见对低资源语言影响更大，因为训练数据有限且文化代表性不足。本文对推理时间偏见缓解进行了全面研究，这是一种避免重新训练或微调的策略，而是直接在模型输出上操作。基于偏好排序模型（PRMs），我们引入了一个统一的评估框架，比较了三种方法：（1）基线单字生成，（2）PRM-Select最佳-N采样，（3）由PRM批评引导的PRM顺序细化。我们在200个英语提示及其乌尔都语对应物上评估这些技术，设计以反映性别、种族、宗教、国籍、残疾、职业、年龄和社会经济类别相关的社会文化背景。使用GPT-3.5作为候选生成器和GPT-4o-mini作为基于PRM的偏见和效用评分器，我们对偏见减少、效用保持和跨语言差异进行了广泛定量分析。我们的发现显示：（a）在两种语言中均优于基线；（b）在所有方法中，乌尔都语的公平性分数始终较低，突显了多语言LLM训练中的结构性不平等；（c）PRM-Select和PRM-顺序细化之间有明显的改进轨迹。本研究贡献了可扩展的方法、可解释的指标和跨语言比较，可支持未来低资源语言公平性评估的工作。
英文摘要：
Large language models (LLMs) increasingly mediate human communication, decision support, content creation, and information retrieval. Despite impressive fluency, these systems frequently produce biased or stereotypical content, especially when prompted with socially sensitive language. A growing body of research has demonstrated that such biases disproportionately affect low-resource languages, where training data is limited and culturally unrepresentative. This paper presents a comprehensive study of inference-time bias mitigation, a strategy that avoids retraining or fine-tuning and instead operates directly on model outputs. Building on preference-ranking models (PRMs), we introduce a unified evaluation framework comparing three methods: (1) baseline single-word generation, (2) PRM-Select best-of-N sampling, and (3) PRM-Sequential refinement guided by PRM critiques. We evaluate these techniques across 200 English prompts and their Urdu counterparts, designed to reflect socio-cultural contexts relevant to gender, ethnicity, religion, nationality, disability, profession, age, and socioeconomic categories. Using GPT-3.5 as a candidate generator and GPT-4o-mini as a PRM-based bias and utility scorer, we provide an extensive quantitative analysis of bias reduction, utility preservation, and cross-lingual disparities. Our findings show: (a) substantial gains over the baseline for both languages; (b) consistently lower fairness scores for Urdu across all methods, highlighting structural inequities in multilingual LLM training; and (c) distinct improvement trajectories between PRM-Select and PRM-Sequential. The study contributes an extensible methodology, interpretable metrics, and cross-lingual comparisons that can support future work on fairness evaluation in low-resource languages.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
