[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84947-en":3,"doc-seo-84947-105":29,"detail-sidebar-cat-0-en-105":90},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":4,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84947,7971461740886,"Theodore","https://ap-avatar.wpscdn.com/davatar_3d24733baf745e90a7e4bdd5f77d97b2",8,"Research & Report","LLMs Silently Correct African American English Auditing and Mitigating Dialect Bias via Activation Steering","African American English (AAE), a rule-governed dialect spoken by over 30 million people, is routinely misinterpreted and “corrected” by large language models. Across six instruction-tuned LLMs (14B to 70B), models systematically prefer Standard American English (SAE) continuations even when prior context is AAE, effectively rewriting AAE into SAE. The work introduces an end-to-end audit and mitigation framework using Dialect Group Invariance and activation steering, plus REALAAE, a large real-AAE parallel corpus.","LLMs Silently Correct African American English: Auditing and Mitigating  \nDialect Bias via Activation Steering  \nHuan Wu 1 ,2 ,3 Ali Emami4 Muhammad Furquan Hassan 1 Osaretin Igbinoba5  \nOsakpolor Idusuyi6 Osamede Igbinoba7 Faiza Khan Khattak8  \nLaleh Seyyed-Kalantari 1 ,2 ,3 ,9 *  \n1 York University 2Vector Institute 3 Connected Minds 4Emory University  \n5Wilfrid Laurier University 6University of Toronto 7University of Guelph  \n8Monark Health 9 CIFAR Solution Network Member  \n∗ [lsk@yorku.ca](lsk@yorku.ca)  \narXiv :2607 .06845v 1 [ cs .CL] 7 Jul 2026  \nAbstract  \nAfrican American English (AAE), a rulegoverned 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 isin 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 REALAAE, the largest real-AAE parallel corpus to date: 17,479 AAE/SAE/AAEback 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) . Our corpus and code are available at [https:](https:)//[anonymous.4open.science/r/dialect](anonymous.4open.science/r/dialect).  \n1 Introduction  \nState-of-the-art large language models (LLMs) silently “correct” African American English (AAE) . Given the AAE context “But I ain’t doing no dishes,”, five out of six LLMs we tested prefer the Standard American English (SAE) continuation  \n* Corresponding author.  \nFigure 1: Dialect preference bias on a real AAE tweet. In a forced-choice continuation task, five out of six state-of-the-art LLMs prefer the SAE continuation over a fluent AAE alternative under an AAE context, and unanimously prefer SAE under an SAE context.  \n“I’m going to stay in my room” over the valid AAE alternative “Imma stay in my room”(Fig. 1) .  \nWe call this behavior dialect preference bias: the systematic tendency of LLMs to favor SAE over valid varieties such as AAE, even when the user is writing in AAE. Treating AAE as something tobe corrected is not a neutral stylistic choice but a form of linguistic discrimination: AAE is a rulegoverned variety with its own consistent grammar and phonology, spoken by over 30 million people (Lin et al., 2025) . The downstream consequences are well-documented: LLMs disproportionately misclassify AAE text (Hassan et al., 2025), respond to AAE speakers with more stereotyping and condescension (Fleisig et al., 2024), and degrade on tasks with non-standard dialect inputs (Ziems et al., 2023 ; Lin et al., 2025) . As LLMs move into hiring, healthcare, and content moderation, this bias actively disadvantages millions of users.  \nExisting work on AAE dialect bias has three key limitations (Table 1) . First, most benchmarks generate AAE synthetically from SAE via rule-based or LLM-driven translation (Ziems et al., 2022, 2023 ; Gupta et al., 2024, 2025), producing AAE that underestimates real-world dialect effects (Lin et al., 2025 ; Gupta et al., 2025) . Second, real-AAE stud-","cbCaipxUWpkD1CNK","https://ap.wps.com/l/cbCaipxUWpkD1CNK","pdf",6059663,1,18,"English","en",105,"# Abstract\n# Introduction\n## Dialect preference bias in LLMs\n## Limitations of existing work\n# Method Overview\n## Auditing with Dialect Group Invariance (cDGI)\n## Feature-level localization of bias triggers\n## Mitigation with activation steering\n# Dataset and Resources\n## REALAAE corpus construction\n## Validation results","[{\"question\":\"What is “dialect preference bias” in the context of this paper?\",\"answer\":\"It is the systematic tendency of LLMs to favor Standard American English (SAE) over valid varieties such as African American English (AAE), even when users write in AAE. The paper shows this preference in forced-choice continuation settings across multiple models.\"},{\"question\":\"How does the paper audit whether bias comes from the model or from translation artifacts?\",\"answer\":\"It introduces conditional Dialect Group Invariance (cDGI), which isolates true model bias from translator-induced artifacts. It also performs feature-level localization to identify which AAE markers most strongly trigger biased behavior.\"},{\"question\":\"How does activation steering mitigate dialect bias according to the paper?\",\"answer\":\"The method is training-free and test-time: it extracts dialect directions via causal tracing and injects them into bias-relevant layers. Results report a 5 to 20× larger bias reduction than prompting while preserving SAE fluency.\"}]",1784199643,45,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":85,"head_meta":87,"extra_data":89,"updated_unix":27},"llms-silently-correct-african-american-english-auditing-and-mitigating-dialect-bias-via-activation-steering","",{"@graph":35,"@context":84},[36,53,67],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/llms-silently-correct-african-american-english-auditing-and-mitigating-dialect-bias-via-activation-steering/84947/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":61,"encodingFormat":60,"isAccessibleForFree":62,"interactionStatistic":63},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-16",true,{"@type":64,"interactionType":65,"userInteractionCount":4},"InteractionCounter",{"@type":66},"ViewAction",{"@type":68,"mainEntity":69},"FAQPage",[70,76,80],{"name":71,"@type":72,"acceptedAnswer":73},"What is “dialect preference bias” in the context of this paper?","Question",{"text":74,"@type":75},"It is the systematic tendency of LLMs to favor Standard American English (SAE) over valid varieties such as African American English (AAE), even when users write in AAE. The paper shows this preference in forced-choice continuation settings across multiple models.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How does the paper audit whether bias comes from the model or from translation artifacts?",{"text":79,"@type":75},"It introduces conditional Dialect Group Invariance (cDGI), which isolates true model bias from translator-induced artifacts. It also performs feature-level localization to identify which AAE markers most strongly trigger biased behavior.",{"name":81,"@type":72,"acceptedAnswer":82},"How does activation steering mitigate dialect bias according to the paper?",{"text":83,"@type":75},"The method is training-free and test-time: it extracts dialect directions via causal tracing and injects them into bias-relevant layers. Results report a 5 to 20× larger bias reduction than prompting while preserving SAE fluency.","https://schema.org",{"og:url":51,"og:type":86,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":88,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":91},[92,96,100,104,109,114,119,122,127,130,134],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":93,"show_sort_weight":94,"slug":95},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":97,"show_sort_weight":98,"slug":99},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":101,"show_sort_weight":102,"slug":103},"Exam",70,"exam",{"id":105,"doc_module":4,"doc_module_name":45,"category_name":106,"show_sort_weight":107,"slug":108},5,"Comic",60,"comic",{"id":110,"doc_module":4,"doc_module_name":45,"category_name":111,"show_sort_weight":112,"slug":113},6,"Technology",50,"technology",{"id":115,"doc_module":4,"doc_module_name":45,"category_name":116,"show_sort_weight":117,"slug":118},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":120,"slug":121},30,"research-report",{"id":123,"doc_module":4,"doc_module_name":45,"category_name":124,"show_sort_weight":125,"slug":126},9,"Religion & Spirituality",20,"religion-spirituality",{"id":125,"doc_module":4,"doc_module_name":45,"category_name":128,"show_sort_weight":125,"slug":129},"World Cup","world-cup",{"id":131,"doc_module":4,"doc_module_name":45,"category_name":132,"show_sort_weight":131,"slug":133},10,"Lifestyle","lifestyle",{"id":135,"doc_module":4,"doc_module_name":45,"category_name":136,"show_sort_weight":105,"slug":137},19,"General","general"]