# LLMs Silently Correct African American English: Auditing and Mitigating Dialect Bias via Activation Steering

arXiv: https://arxiv.org/abs/2607.06845 · HTML: https://mlsanity.com/paper/2607.06845/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Probing deception, bias and beliefs, one of 13 directions in Mechanistic interpretability.
- Builds on 19 of its 31 references that are in the map.
- First posted on arXiv 7 Jul 2026.

## Placed in

- Mechanistic interpretability / [Probing deception, bias and beliefs](https://mlsanity.com/mechinterp/probing-deception-bias-and-beliefs/)

## Builds on (most cited first, 19 of 19)

- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Phi-3 Technical Report: A Highly Capable Language Model Locally on Your Phone](https://mlsanity.com/paper/2404.14219/) (2024)
- [Steering Llama 2 via Contrastive Activation Addition](https://mlsanity.com/paper/2312.06681/) (2023)
- [Refusal in Language Models Is Mediated by a Single Direction](https://mlsanity.com/paper/2406.11717/) (2024)
- [Persona Vectors: Monitoring and Controlling Character Traits in Language Models](https://mlsanity.com/paper/2507.21509/) (2025)
- [Annotators with Attitudes: How Annotator Beliefs And Identities Bias Toxic Language Detection](https://mlsanity.com/paper/2111.07997/) (2021)
- [In-context Vectors: Making In Context Learning More Effective and Controllable Through Latent Space Steering](https://mlsanity.com/paper/2311.06668/) (2023)
- [Racial Bias in Hate Speech and Abusive Language Detection Datasets](https://mlsanity.com/paper/1905.12516/) (2019)
- [Linguistic Bias in ChatGPT: Language Models Reinforce Dialect Discrimination](https://mlsanity.com/paper/2406.08818/) (2024)
- [Multi-VALUE: A Framework for Cross-Dialectal English NLP](https://mlsanity.com/paper/2212.08011/) (2022)
- [VALUE: Understanding Dialect Disparity in NLU](https://mlsanity.com/paper/2204.03031/) (2022)
- [Evaluation of African American Language Bias in Natural Language Generation](https://mlsanity.com/paper/2305.14291/) (2023)
- [Assessing Dialect Fairness and Robustness of Large Language Models in Reasoning Tasks](https://mlsanity.com/paper/2410.11005/) (2024)
- [Rejected Dialects: Biases Against African American Language in Reward Models](https://mlsanity.com/paper/2502.12858/) (2025)
- [AAVENUE: Detecting LLM Biases on NLU Tasks in AAVE via a Novel Benchmark](https://mlsanity.com/paper/2408.14845/) (2024)
- [Predicting the Target Word of Game-playing Conversations using a Low-Rank Dialect Adapter for Decoder Models](https://mlsanity.com/paper/2409.00358/) (2024)
- [EnDive: A Cross-Dialect Benchmark for Fairness and Performance in Large Language Models](https://mlsanity.com/paper/2504.07100/) (2025)
- [DialectGen: Benchmarking and Improving Dialect Robustness in Multimodal Generation](https://mlsanity.com/paper/2510.14949/) (2025)

## Abstract (arXiv)

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).