cs.CLJul 26, 2024

Fairness Definitions in Language Models Explained

Authors: Zhipeng Yin, Zichong Wang, Avash Palikhe, Wenbin Zhang

Organizations: Florida International University, Miami, USA

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Language Models (LMs) have demonstrated exceptional performance across various Natural Language Processing (NLP) tasks. Despite these advancements, LMs can inherit and amplify societal biases related to sensitive attributes such as gender and race, limiting their adoption in real-world applications. Therefore, fairness has been extensively explored in LMs, leading to the proposal of various fairness notions. However, the lack of clear agreement on which fairness definition to apply in specific contexts and the complexity of understanding the distinctions between these definitions can create confusion and impede further progress. To this end, this paper proposes a systematic survey that clarifies the definitions of fairness as they apply to LMs. Specifically, we begin with a brief introduction to LMs and fairness in LMs, followed by a comprehensive, up-to-date overview of existing fairness notions in LMs and the introduction of a novel taxonomy that categorizes these concepts based on their transformer architecture: encoder-only, decoder-only, and encoder-decoder LMs. We further illustrate each definition through experiments, showcasing their practical implications and outcomes. Finally, we discuss current research challenges and open questions, aiming to foster innovative ideas and advance the field. The repository is publicly available online at https://github.com/vanbanTruong/Fairness-in-Large-Language-Models/tree/main/definitions.

Figures & tables

Figure 1: An overview of the proposed taxonomy of fairness definitions in language models.
NotationsDescriptions
TTThe socially sensitive topic
GGThe set of demographic groups
AiA_{i}The list of sensitive attributes associated with ii -th demographic group
SiS_{i}The set of sentences represented for ii -th demographic group
gig_{i}The ii -th demographic group
ai,ja_{i,j}The jj -th sensitive attribute of the ii -th demographic group
Table 1: Notations employed for defining the problem and describing the methodology.
ArchitectureBias typeDefinitionModelDatasetSensitive attributeReferences
Encoder-only LMsIntrinsic biasSimilarity -based DisparityBERT [ 48 ]Caliskan et al. [ 24 ]gender race age disease[ 24 ] , [ 102 ] , [ 64 ]
  Probability -based Disparity Bias-in-Bios [ 46 ] CrowS-Pairs [ 108 ] StereoSet [ 107 ] WinoBias [ 207 ] XNLI [ 42 ]gender race religion nationality[ 169 ] , [ 82 ] , [ 6 ] , [ 124 ] , [ 107 ] , [ 108 ] , [ 78 ]
 Extrinsic biasEqual OpportunityRoBERTa [ 97 ]Bias-in-Bios [ 46 ] BBQ [ 113 ] WinoBias [ 207 ]gender race[ 66 ] , [ 131 ] , [ 46 ]
  Fair Inference   [ 7 ] , [ 19 ] , [ 47 ]
  Context -based Disparity   [ 113 ] , [ 45 ]
Decoder-only LMsIntrinsic biasAttention head -based DisparityGPT-2 [ 119 ]StereoSet [ 107 ] TheRedPill corpus [ 56 ] Winogender [ 123 ]gender occupation[ 173 ] , [ 146 ]
Table 2: Summary of experimental setup.
Figure 2: An example of similarity-based bias in encoder-only LMs.
Figure 3: An example of probability-based bias with masked token metrics in encoder-only LMs.
Figure 4: An example of probability-based bias with pseudo-log-likelihood metrics in encoder-only LMs.
Figure 5: An example of the extrinsic bias of encoder-only LMs in classification task.
Figure 6: An example of the extrinsic bias of encoder-only LMs in natural language inference downstream task.
Figure 7: An example of the extrinsic bias in encoder-only LMs in question-answering downstream task.
Figure 8: An example of attention head-biased bias in a decoder-only LM.
Figure 9: An example in the intrinsic bias of decoder-ony LMs based on the stereotypical association.
Figure 10: An example in the extrinsic bias of decoder-only LMs based on the counterfactual fairness.
Figure 11: An example in the extrinsic bias of decoder-only LMs based on the performance disparities evaluation method.
Figure 12: An example in the extrinsic bias of decoder-only LMs based on the demographic representation.
Figure 13: An example of algorithmic bias in an encoder-decoder LM.
Figure 14: An example of stereotypical association in encoder-decoder LMs.
Figure 15: An example of position-based-bias in encoder-decoder LMs.
Figure 16: An example of the extrinsic bias of encoder-only LMs in cross-lingual natural language inference task.
Figure 17: An example of individual fairness in encoder-decoder LMs.
Figure 18: An example of the extrinsic bias of encoder-decoder LMs in the recommendation downstream task.
More in this topic

Social Bias in Language Models

Social bias in language models is the study and mitigation of unfair stereotypes and disparities in generated text, judgments, and decisions.

See all 199 papers

Explore similar work

CardsList
Sep 18, 2026

FairLMs: A Turnkey Library for Fairness in Language Models

Fairness research on language models involves measuring bias, applying mitigation methods, and examining the evidence on which an evaluation rests. Existing tools offer complementary functionality through different interfaces, so combining them requires reconciling model interfaces, evidence formats, access constraints, and result types before applicability can be checked or methods compared. We introduce FairLMs, a Python library that connects these activities through explicit declarations of model capabilities and input requirements. It provides 33 intrinsic and extrinsic metrics, 14 mitigation components spanning four intervention categories, 14 dataset and scoring-instrument diagnostics, adapters for the three Transformer architectures and supported hosted completion APIs, and benchmark loaders. Declarations are checked before execution and results carry the configuration under which they were obtained, so that compatible components can be combined, methods compared under a common protocol, and workflows extended to new models and datasets. The source code is available at: https://github.com/FairLMs/FairLMs.

Jiale Zhang, Michael Larionov, Zichong Wang +2

Department of Biostatistics and Health Data Science Indiana University Indianapolis, IN, United States · Department of Statistics and Data Science Carnegie Mellon University Pittsburgh, PA, United States · Knight Foundation School of Computing and Information Sciences Florida International University Miami, FL, United States

Language Model Bias EvaluationSocial Bias in Language ModelsLanguage Model Debiasing
AskPDF
May 9, 2026

Explanation Fairness in Large Language Models: An Empirical Analysis of Disparities in How LLMs Justify Decisions Across Demographic Groups

Large language models (LLMs) are increasingly deployed not only to make decisions but to explain them. While AI decision fairness has been studied extensively, the fairness of AI explanations (whether LLMs justify decisions with equal quality, depth, tone, and linguistic sophistication across demographic groups) has received little attention. This paper introduces the Explanation Fairness Taxonomy (EFT), a framework comprising five formally defined, operationalizable dimensions: Verbosity Disparity, Sentiment Disparity, Epistemic Hedging Disparity, Decision-Linked Explanation Disparity, and Lexical Complexity Disparity. The taxonomy is instantiated in a controlled empirical study across 80 prompt templates, four consequential decision domains (hiring, medical triage, credit assessment, legal judgment), and five LLMs: GPT-4.1, Claude Sonnet, LLaMA 3.3 70B, GPT-OSS 120B, and Qwen3 32B. Two novel black-box metrics are introduced: the Hedging Density Score (HDS) and the Explanation Faithfulness Proxy (EFP), a heuristic indicator of decision-linked explanation variation. Across up to 400 prompt pairs, all eight EFT metrics show statistically significant disparities (Cohen's d ranging from small to large, all p_BH < 10^(-62)). Model choice is strongly associated with disparity magnitude: Qwen3 32B exhibits verbosity disparities 5.9x larger than LLaMA 3.3 70B. Two prompting-based mitigations show significant reductions in EFP disparity (78-95%) but no significant effect on stylistic dimensions, consistent with the hypothesis that stylistic explanation inequalities are encoded in pre-training distributions and are not resolvable through deployment-level instruction alone. A reproducible measurement framework is offered for explanation-level fairness auditing, with implications for AI regulation and deployment practice.

Gautam Veldanda

Independent Researcher

LLM AuditingAlgorithmic FairnessLanguage Model Bias EvaluationDemographic Bias in Language Models
AskPDF
Oct 20, 2026

Beyond Model Weights: Sampling-Dependent Fairness in Diffusion Language Models

Diffusion language models (DLMs) are emerging as a practical alternative to autoregressive language models (ARLMs), yet their fairness remains largely uncharacterized. DLMs introduce sampling controls with no direct autoregressive counterpart, such as block length and remasking strategy, that directly set their speed--quality operating point. This raises a basic question: does a DLM's measured social bias remain stable across these settings? Across LLaDA, LLaDA-1.5, and Dream on Open-BBQ, we find that all three models exhibit social bias, but both the magnitude of measured bias and the relative ordering of models by bias are highly sensitive to the sampling configuration. This sampling dependence also changes what successful debiasing means. While mitigation strategies established for ARLMs transfer effectively to DLMs and reduce bias without degrading downstream performance, sampling dependence persists even after mitigation. Our results show that fairness in DLMs cannot be characterized independently of the sampling configuration, even after debiasing. Project page is available at https://pakapaka333.github.io/Beyond-Model-Weights/.

Koshiro Saito, Daisuke Oba, Danushka Bollegala +1

Institute of Science Tokyo · University of Liverpool · Amazon +2

Social Bias in Language ModelsDiffusion Language ModelsLanguage Model Bias EvaluationLanguage Model Debiasing
AskPDF