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

Huan Wu submitted a paper to arXiv on July 7, 2026, titled "LLMs Silently Correct African American English." The research audits and mitigates dialect bias in large language models through activation steering, examining how LLMs inadvertently alter African American English outputs. The paper, co-authored with seven other contributors, falls under the cs.CL category and is accessible via PDF and HTML formats on the arXiv platform.