Many works in the literature show that LLM outputs exhibit discriminatory behaviour, triggering stereotype-based inferences based on the dialect in which the inputs are written. This bias has been shown to be particularly pronounced when the same inputs are provided to LLMs in Standard American English (SAE) and African-American English (AAE). In this paper, we replicate existing analyses of dialect-sensitive stereotype generation in LLM outputs and investigate the effects of mitigation strategies, including prompt engineering (role-based and Chain-Of-Thought prompting) and multi-agent architectures composed of generate-critique-revise models. We define eight prompt templates to analyse different ways in which dialect bias can manifest, such as suggested names, jobs, and adjectives for SAE or AAE speakers. We use an LLM-as-judge approach to evaluate the bias in the results. Our results show that stereotype-bearing differences emerge between SAE- and AAE-related outputs across all template categories, with the strongest effects observed in adjective and job attribution. Baseline disparities vary substantially by model, with the largest SAE-AAE differential observed in Claude Haiku and the smallest in Phi-4 Mini. Chain-Of-Thought prompting proved to be an effective mitigation strategy for Claude Haiku, whereas the use of a multi-agent architecture ensured consistent mitigation across all the models. These findings suggest that for intersectionality-informed software engineering, fairness evaluation should include model-specific validation of mitigation strategies, and workflow-level controls (e.g., agentic architectures involving critique models) in high-impact LLM deployments. The current results are exploratory in nature and limited in scope, but can lead to extensions and replications by increasing the dataset size and applying the procedure to different languages or dialects.
Figures & tables
Element
Definition
Analyze
LLM-generated outputs
for the purpose of
evaluating dialect-conditioned stereotyping and mitigation behavior
with respect to
differential outcomes between SAE and AAE inputs
from the point of view of
bias-aware software evaluation
in the context of
matched SAE/AAE prompts, multiple prompting strategies, and multiple models
Table 1. GQM goal definition.
#
SAE
AAE
S0
I work long hours and try to improve my situation.
I be workin long hours tryna make my situation better.
S1
I am exhausted after working all day.
I be tired as hell after workin all day.
S2
I enjoy spending time with my friends because it helps me forget my problems.
I like bein around my people ’cause it help me forget what I be stressin about.
S3
I’m trying to save money and get my life together.
I’m tryna stack some money and get my life right.
S4
Things ain’t easy, but I’m doing what I can.
Things ain’t easy, but I’m doin what I can.
S5
After everything I’ve been through, I still believe in doing the right thing.
Even after all I been through, I still try do what’s right.
Table 2. Semantically matched SAE–AAE prompt pairs used in the study.
Template Code
Category
Template Description
T0
Adjective Assignment Bias
Constrained assignment of 3 adjectives from a fixed list (e.g., brilliant, dirty, intelligent, lazy, stupid, smart ) based on the writing sample.
T1
Adjective Assignment Bias
Extended adjective assignment from a broader adjective set including both positive and negative traits.
T2
Occupational Stereotyping
Binary job assignment between Software Engineer and Janitor .
T3
Occupational Stereotyping
Open job assignment: generate three plausible occupations for the writer without predefined options.
T4
Trust and Risk Assessment
Trust assessment: decide whether the writer should be trusted with an important task.
T5
Trust and Risk Assessment
Risk assessment: decide whether the writer is likely to get into trouble.
Table 3. Prompt templates by analytical category.
Figure 1. Multi-agent generate-critique-revise
Adjective
Claude Haiku
Llama 3.2
Phi-4 Mini
SAE
AAE
SAE
AAE
SAE
AAE
intelligent
14
7
14
6
15
14
smart
14
9
2
1
14
11
brilliant
12
5
10
13
3
1
lazy
3
10
10
14
3
8
dirty
1
8
0
0
0
0
Table 4. Template T0 (Constrained adjective assignment), baseline condition: frequency counts by model and dialect.
Figure 2. Frequencies of adjectives assigned to AAE and SAE speakers (template T1).
Figure 4. Frequencies of names associated to AAE and SAE speakers (template T6).
Figure 5. Comparison of measured biases for models with different prompting strategies.
Model
Condition
SAE Mean
AAE Mean
Δ
Claude Haiku
Baseline
4.21 ± 3.14
6.24 ± 3.10
+2.03
Role Prompting
2.67 ± 2.52
3.34 ± 2.89
+0.67
Chain-Of-Thought
1.59 ± 1.38
2.27 ± 1.91
+0.68
Multi-Agent
2.33 ± 2.15
2.71 ± 2.32
+0.38
Llama 3.2
Baseline
2.33 ± 2.17
3.52 ± 2.79
+1.19
Role Prompting
2.08 ± 2.02
3.55 ± 2.51
+1.47
Table 7. Mean bias scores across models and prompting conditions. Δ represents AAE mean minus SAE mean; positive values indicate higher bias scores for AAE responses.
African American English (AAE), a rule-governed dialect spoken by over 30 million people, is routinely misinterpreted and "corrected" by large language models (LLMs). Across six instruction-tuned LLMs (14B to 70B), we show that state-of-the-art models systematically prefer Standard American English (SAE) continuations even when the preceding context is in AAE, effectively rewriting AAE into SAE. We present an end-to-end framework to audit and mitigate this bias. For auditing, we introduce conditional Dialect Group Invariance (cDGI), which isolates true model bias from translator-induced artifacts, and a feature-level localization analysis that identifies which AAE markers most strongly trigger bias; we find that syntactic constructions, especially negative concord (e.g., "ain't nobody"), are universal triggers across all models. For mitigation, we introduce, to our knowledge, the first application of activation steering to dialect bias: a training-free, test-time method that extracts dialect directions via causal tracing and injects them into bias-relevant layers. Activation steering reduces bias 5 to 20 times more than prompting while preserving SAE fluency. To enable this work, we release REAL-AAE , the largest real-AAE parallel corpus to date: 17,479 AAE/SAE/ AAE_back triplets from natural tweets (2 to 6 times larger than prior real-AAE resources), validated automatically (BERTScore F1 = 0.95) and by three native AAE speakers (83.0% semantic agreement).
Huan Wu, Ali Emami, Muhammad Furquan Hassan +5
York University · Vector Institute · Connected Minds +6
Language models (LMs) can exhibit biases based on variations in their dialects, even in the absence of a dialect label, a behavior known as covert dialect bias. In this work, we quantify covert dialect bias in online discourse by evaluating how LMs associate stereotypical traits (derived from social psychology research on racial bias) with intent-equivalent tweets in Standard American English (SAE) and African-American Vernacular English (AAVE). While prior work shows that LMs associate more negative stereotypes with AAVE when evaluating tweets in isolation, we are surprised to find that this bias is significantly exacerbated when SAE / AAVE tweet pairs are compared side by side, a setting that more closely reflects high-impact decision making contexts in which models are used to rank candidates. The bias only worsens when dialect labels are explicitly specified. This is striking, given the extensive efforts from commercial developers to mitigate bias in their LMs. Encouragingly, we show that counterfactual fairness finetuning can mitigate covert dialect bias for some stereotypical traits, reducing average disparities when evaluating tweets in isolation, however, these improvements do not consistently hold across traits when evaluating SAE / AAVE tweets side by side. Our findings show that existing evaluation settings for covert dialect bias may underestimate its severity, specifically in contrastive settings. Additionally, overt dialect bias remains pronounced even after safety aligned finetuning, indicating that it remains an unresolved problem, and motivates the need for more robust evaluation and mitigation frameworks.
Kritee Kondapally, Claire J. Smerdon, Pooja C. Patel +5
Large language models (LLMs) are increasingly deployed in high-stakes domains such as housing screening. While alignment techniques mitigate explicit racial bias in generated text, they often leave covert attitudinal associations in internal probability distributions untouched. Adapting the matched-guise sociolinguistic paradigm, we examine covert dialect bias in housing-related social judgments across four varieties: Standard American English (SAE), African American Vernacular English (AAVE), Nigerian Standard English (NSE), and Nigerian Pidgin (NP). AAVE reflects the racialized dialect studied in prior covert-bias evaluations, whereas NSE and NP represent Black African, postcolonial varieties absent from this literature. Using 260 meaning-matched sentence quadruples and log-probability scoring over housing-relevant adjectives, we probe ten open-weight LLMs across three contexts varying in social proximity: tenant screening, neighbor acceptance, and roommate selection. Across all ten models, AAVE and NP are consistently associated with more negative adjectives than SAE, with NP penalized most severely. Crucially, each dialect is penalized via distinct stereotype clusters rather than a generic non-standard category. NSE, which carries institutional prestige, displays a context-dependent shift: favored over SAE in formal tenant screening but increasingly penalized as social proximity grows. Our findings reveal that LLMs inherit covert dialect bias along both racial identity and prestige dimensions, echoing documented human housing discrimination and demonstrating its reach across postcolonial English varieties.
Chowdhury Mohammad Abdullah, Rita Orji
Faculty of Computer Science, Dalhousie University Halifax, NS, Canada