[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-83316-en":3,"doc-seo-83316-105":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},83316,1374391974585,"Genevieve","https://ap-avatar.wpscdn.com/davatar_276721f389ce27ea32af1340a28f341c",8,"Research & Report","PLURAL: A Global Dataset for Value Alignment","Large language models are widely deployed yet often mirror Western value systems, reducing their ability to represent diverse cultural norms. PLURAL introduces a large-scale, value-focused preference dataset grounded in the Integrated Values Survey across 92 countries. A two-stage generation pipeline converts survey answers into synthetic preference triplets that retain normative value signals while producing realistic scenarios. An initial release includes ~500,000 triplets from 20 countries, validated at dataset, automated, and blind human evaluation levels.","arXiv :2607 .08034v 1 [ cs .CL] 9 Jul 2026  \nPLURAL: A Global Dataset for Value Alignment  \nDhruv AgarwalAnya Shukla, Tanya Goyal & Aditya Vashistha Cornell University  \nAbstract  \nLarge language models (LLMs) are used worldwide, yet disproportionately reflect Western values, limiting their ability to represent diverse value systems. We introduce PLURAL, a large-scale, value-focused preference dataset grounded in the Integrated Values Survey (IVS), a nationally representative survey spanning 92 countries. Using a two-stage generation pipeline, we transform survey responses into synthetic preference triplets that preserve normative value signals while producing realistic scenarios.  \nWe release an initial version of PLURAL containing ∼500,000 preference triplets representing people in 20 diverse countries. We evaluate PLURAL in three ways: (i) dataset-level validation showing that it preserves both cross-country value differences and within-country diversity from the original survey; (ii) automated evaluation showing that training on PLURAL improves alignment with target countries’ cultural profiles, reducing mean absolute error by up to 27.7% relative to strong baselines; and (iii) blind human evaluation with 176 evaluators in India, Brazil, and Japan, who judge PLURAL-aligned responses as more representative of their national values. Together, these results show that PLURAL contains learnable signal for value steering, offering a scalable resource for pluralistic alignment.  \n [Dataset:](Dataset: huggingface.co/datasets/agdhruv/plural-alignment)[ huggingface.co/datasets/agdhruv/plural-alignment](Dataset: huggingface.co/datasets/agdhruv/plural-alignment)  \n1 Introduction  \nLarge Language Models (LLMs) increasingly mediate everyday tasks from writing assistance to complex decision-making, raising concerns about whose values they reflect (Prabhakaran et al., 2022) . A growing body of work shows that these models disproportionately encode Western cultures and values (Johnson et al., 2022; Cao et al., 2023; Qadri et al., 2023) . This Western-centric anchoring can produce tangible harms for international users, such as homogenizing values during writing (Agarwal et al., 2025a; Shahid et al., 2026) .  \nTo address these representational harms, the AI community has increasingly focused on developing regional or “sovereign” language models tailored to local languages and communities, such as Nanda for Hindi (Choudhury et al., 2025) and Jais for Arabic (Sengupta et al., 2023) . Yet, recent work shows that regional models are often “fluent but foreign”: they learn local syntax yet continue to reflect American cultural values, highlighting that achieving linguistic fluency in a target language does not guarantee value alignment (Agarwal et al., 2025b) . Thus, the challenge is not merely to build models that speak different languages, but to build models that can engage with the plurality of human values across cultures, abroader problem formalized as pluralistic alignment (Sorensen et al., 2024) .  \nExisting alignment efforts typically rely on preference-learning methods such as RLHF (Ouyang et al., 2022) and DPO (Rafailov et al., 2024), which depend on human preference data to steer model behavior. In practice, however, such data have historically been collected disproportionately from Western, English-speaking populations, thus optimizing models for relatively narrow demographic slices (Bai et al., 2022) . More recent efforts such as PRISM and Community Alignment broaden representation, but important limitations remain. PRISM spans 75 countries but remains heavily skewed toward Western and highly educated respondents, with over 65% white and the US/UK contributing over  \n∗ Correspondence: {da399,[adityav](adityav}@cornell.edu)[}](adityav}@cornell.edu)[@cornell.edu](adityav}@cornell.edu)  \nFigure 1: Overview of the PLURAL generation and evaluation pipeline.  \n40% of participants (Kirk et al., 2024) . Community Alignment improves on these","cbCaiehtGIeSpsBy","https://ap.wps.com/l/cbCaiehtGIeSpsBy","pdf",2574453,1,27,"English","en",105,"# Introduction\n## Motivation: Western-centric value encoding\n## Related work and limitations in preference data\n## PLURAL dataset construction and usability","[{\"question\":\"What problem does PLURAL address?\",\"answer\":\"PLURAL addresses the tendency of large language models to reflect Western values disproportionately, which limits representation of diverse value systems for international users.\"},{\"question\":\"How is the PLURAL dataset created?\",\"answer\":\"PLURAL is grounded in the Integrated Values Survey and uses a two-stage pipeline to transform survey responses into synthetic preference triplets that preserve normative value signals.\"},{\"question\":\"How does PLURAL get evaluated and what results does it show?\",\"answer\":\"PLURAL is evaluated through dataset-level validation, automated evaluation using country-specific training, and blind human evaluation with evaluators from India, Brazil, and Japan, showing improved alignment to target cultural profiles and better national-value representativeness.\"}]",1784186695,68,{"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},"plural-a-global-dataset-for-value-alignment","",{"@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/plural-a-global-dataset-for-value-alignment/83316/",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 problem does PLURAL address?","Question",{"text":74,"@type":75},"PLURAL addresses the tendency of large language models to reflect Western values disproportionately, which limits representation of diverse value systems for international users.","Answer",{"name":77,"@type":72,"acceptedAnswer":78},"How is the PLURAL dataset created?",{"text":79,"@type":75},"PLURAL is grounded in the Integrated Values Survey and uses a two-stage pipeline to transform survey responses into synthetic preference triplets that preserve normative value signals.",{"name":81,"@type":72,"acceptedAnswer":82},"How does PLURAL get evaluated and what results does it show?",{"text":83,"@type":75},"PLURAL is evaluated through dataset-level validation, automated evaluation using country-specific training, and blind human evaluation with evaluators from India, Brazil, and Japan, showing improved alignment to target cultural profiles and better national-value representativeness.","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"]