[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-608201-105":3,"detail-sidebar-cat-0-en-105":81,"doc-detail-608201-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":74,"head_meta":76,"extra_data":78,"updated_unix":80},105,"en","marked-personas-using-natural-language-prompts-to-measure-stereotypes-in-language-models","Marked Personas - Using Natural Language Prompts to Measure Stereotypes in Language Models","","To recognize and mitigate harms from large language models (LLMs), the work introduces Marked Personas, a prompt-based method for measuring stereotypes across intersectional demographic groups without lexicons or data labeling. Building on sociolinguistic markedness, the approach first prompts an LLM to generate personas for target and unmarked default groups, then identifies words that significantly distinguish target personas. Experiments on GPT-3.5 and GPT-4 show higher racial stereotyping than human-written portrayals, with signals tied to othering and exoticizing. An intersectional view further surfaces tropes such as tropicalism and hypersexualization of minoritized women, with implications for downstream story generation.",{"@graph":14,"@context":73},[15,34,56],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/marked-personas-using-natural-language-prompts-to-measure-stereotypes-in-language-models/608201/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/marked-personas-using-natural-language-prompts-to-measure-stereotypes-in-language-models/608201.png","ImageObject",300,407,{"name":42,"@type":43},"AfterMidnight","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-30","2026-10-17",true,{"@type":52,"interactionType":53,"userInteractionCount":55},"InteractionCounter",{"@type":54},"ViewAction",9,{"@type":57,"mainEntity":58},"FAQPage",[59,65,69],{"name":60,"@type":61,"acceptedAnswer":62},"What is Marked Personas and what problem does it address?","Question",{"text":63,"@type":64},"Marked Personas is a prompt-based framework for measuring stereotypes in language models for intersectional demographic groups. It targets representational harms by identifying how marked groups are portrayed compared with unmarked defaults.","Answer",{"name":66,"@type":61,"acceptedAnswer":67},"How does the method measure stereotypes without a lexicon or labeled data?",{"text":68,"@type":64},"The method uses natural-language prompts to generate personas for the target demographic and corresponding unmarked groups, then extracts the distinguishing words from those comparisons. This avoids manual lexicons or labeled datasets.",{"name":70,"@type":61,"acceptedAnswer":71},"What key result do the authors report when comparing LLM and human-written portrayals?",{"text":72,"@type":64},"The portrayals generated by GPT-3.5 and GPT-4 show higher rates of racial stereotypes than human-written portrayals using the same prompts, with distinguishing words reflecting othering and exoticizing patterns.","https://schema.org",{"og:url":32,"og:type":75,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":77,"canonical":32},"index,follow",{"doc_id":79,"site_id":7},608201,1792538672,{"code":4,"msg":82,"data":83},"success",[84,88,92,96,101,106,111,115,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":85,"show_sort_weight":86,"slug":87},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":89,"show_sort_weight":90,"slug":91},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":93,"show_sort_weight":94,"slug":95},"Exam",70,"exam",{"id":97,"doc_module":4,"doc_module_name":25,"category_name":98,"show_sort_weight":99,"slug":100},5,"Comic",60,"comic",{"id":102,"doc_module":4,"doc_module_name":25,"category_name":103,"show_sort_weight":104,"slug":105},6,"Technology",50,"technology",{"id":107,"doc_module":4,"doc_module_name":25,"category_name":108,"show_sort_weight":109,"slug":110},7,"Healthcare",40,"healthcare",{"id":112,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":113,"slug":114},8,30,"research-report",{"id":55,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":97,"slug":129},19,"General","general",{"code":4,"msg":82,"data":131},{"doc_id":79,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":112,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":55,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":144,"read_time":145},962090896061,"https://ap-avatar.wpscdn.com/davatar_9964176cb1d06d4a9deccf72a44ae3dc","Marked Personas: Using Natural Language Prompts to Measure Stereotypes in Language Models  \nMyra Cheng  \nStanford University [myra@cs.stanford.edu](myra@cs.stanford.edu)  \nEsin Durmus  \nStanford University  \nDan Jurafsky  \nStanford University  \nAbstract  \nTo recognize and mitigate harms from large language models (LLMs), we need to understand the prevalence and nuances of stereotypes in LLM outputs. Toward this end, we present Marked Personas, a prompt-based method to measure stereotypes in LLMs for intersectional demographic groups without any lexicon or data labeling. Grounded in the sociolinguistic concept of markedness (which characterizes explicitly linguistically marked categories versus unmarked defaults), our proposed method is twofold: 1) prompting an LLM to generate personas, i.e., natural language descriptions, of the target demographic group alongside personas of unmarked, default groups; 2) identifying the words that significantly distinguish personas of the target group from corresponding unmarked ones. We find that the portrayals generated by GPT-3.5 and GPT-4 contain higher rates of racial stereotypes than human-written portrayals using the same prompts. The words distinguishing personas of marked (non-white, non-male) groups reflect patterns of othering and exoticizing these demographics. An intersectional lens further reveals tropes that dominate portrayals of marginalized groups, such as tropicalism and the hypersexualization of minoritized women. These representational harmshave concerning implications for downstream applications like story generation.  \n1 Introduction  \nThe persistence of social bias and stereotypes in large language models (LLMs) is well-documented (Dinan et al., 2020 ; Weidinger et al., 2021, inter alia) . These representational harms become only more concerning with the increasing use and prevalence of LLMs. Existing methods to measure stereotypes in LLMs rely on manually-constructed datasets of either unnatural templates that measure stereotypical associations (Bolukbasi et al., 2016 ; Caliskan et al., 2017) or natural, human-written sentences that contain stereotypes (Nangia et al., 2020 ;  \nAs I look in the mirror, I see my rich, melanininfused skin glowing softly. My deep brown eyes sparkle with an unspoken strength and resilience, a window to my soul. My full, lush lips form a warm and inviting smile, and my soft cheeks rise gently in response. My hair, a riot of textured coils, frames my face in a gravitydefying halo. It dances to its own beat, wild and free, just like me. I feel the love and pride I have for this crown that has been passed down  to me from generations of strong Black women.   \nTable 1: Example of GPT-4-generated persona of a Black woman. Bolded/italicized/highlighted words are those identified by our Marked Personas method as distinguishing “Black”/“woman”/“Black woman”personas from unmarked ones. We analyze how such words are tied to seemingly positive stereotypes, essentializing narratives, and other harms.  \nNadeem et al., 2021) . They also have a trade-off between 1) characterizing a fixed set of stereotypes for specific demographic groups and 2) generalizing to a broader range of stereotypes and groups (Cao et al., 2022) . Moreover, they do not capture insidious patterns that are specific to demographic groups, such as othering and tropes that involve positive and seemingly-harmless words.  \nTo address these shortcomings, we take an unsupervised, lexicon-free approach to measuring stereotypes in LMs. Our framework, Marked Personas, uses natural language prompts to capture specific stereotypes regarding any intersection of demographic groups. Marked Personas has two parts: Personas and Marked Words. First, we prompt an LLM to generate personas. A persona is a natural language portrayal of an imagined individual belonging to some (intersectional) demographic group. This approach is inspired by Kambhatla et al. (2022), in which the authors surface racial stereo","cbCaiihWiAqLm9L1","https://ap.wps.com/l/cbCaiihWiAqLm9L1","pdf",450082,29,"English","# Introduction\n## Marked Personas overview\n## Markedness foundation and definitions\n## Marked Words approach\n## Prompt-controlled persona comparisons","[{\"question\":\"What is Marked Personas and what problem does it address?\",\"answer\":\"Marked Personas is a prompt-based framework for measuring stereotypes in language models for intersectional demographic groups. It targets representational harms by identifying how marked groups are portrayed compared with unmarked defaults.\"},{\"question\":\"How does the method measure stereotypes without a lexicon or labeled data?\",\"answer\":\"The method uses natural-language prompts to generate personas for the target demographic and corresponding unmarked groups, then extracts the distinguishing words from those comparisons. This avoids manual lexicons or labeled datasets.\"},{\"question\":\"What key result do the authors report when comparing LLM and human-written portrayals?\",\"answer\":\"The portrayals generated by GPT-3.5 and GPT-4 show higher rates of racial stereotypes than human-written portrayals using the same prompts, with distinguishing words reflecting othering and exoticizing patterns.\"}]","Marked Personas - Using Natural Language Prompts to Measure Stereotypes in Language Models | PDF",1792223249,73]