Responsible AI in NLP: GUS-Net Span-Level Bias Detection Dataset and Benchmark for Generalizations, Unfairness, and Stereotypes

Responsible AI in NLP: GUS-Net Span-Level Bias Detection Dataset and Benchmark for Generalizations, Unfairness, and Stereotypes | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Responsible AI in NLP: GUS-Net Span-Level Bias Detection Dataset and Benchmark for Generalizations, Unfairness, and Stereotypes Maximus Powers, Shaina Raza, Alex Chang, Rehana Riaz, Umang Mavani, and 3 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-7623811/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Representational harms in language technologies often occur in short spans within otherwise neutral text, where phrases may simultaneously convey generalizations, unfairness, or stereotypes. Framing bias detection as sentence-level classification obscures which words carry bias and what type is present, limiting both auditability and targeted mitigation. We introduce the GUS-Net Framework, comprising the GUS dataset and a multi-label token-level detector for span-level analysis of social bias. The GUS dataset contains 3,739 unique snippets across multiple domains, with over 69,000 token-level annotations. Each token is labeled using BIO tags (Begin, Inside, Outside) for three pathways of representational harm: Generalizations, Unfairness, and Stereotypes. To ensure reliable data annotation, we employ an automated multi-agent pipeline that proposes candidate spans which are subsequently verified and corrected by human experts. We formulate bias detection as multi-label token-level classification and benchmark both encoder-based models (e.g., BERT family variants) and decoder-based large language models (LLMs). Our evaluations cover token-level identification and span-level entity recognition on our test set, and out-of-distribution generalization. Empirical results show that encoder-based models consistently outperform decoder-based baselines on nuanced and overlapping spans while being more computationally efficient. The framework delivers interpretable, fine-grained diagnostics that enable systematic auditing and mitigation of representational harms in real-world NLP systems. Bias detection Token-level tagging BIO Social bias Responsible AI Dataset NLP Full Text Additional Declarations No competing interests reported. Supplementary Files AuthorshipContributionStatement.docx Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 11 May, 2026 Reviewers agreed at journal 23 Apr, 2026 Reviews received at journal 30 Mar, 2026 Reviewers agreed at journal 17 Mar, 2026 Reviewers agreed at journal 03 Oct, 2025 Reviewers invited by journal 01 Oct, 2025 Editor assigned by journal 25 Sep, 2025 Submission checks completed at journal 16 Sep, 2025 First submitted to journal 15 Sep, 2025 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-7623811","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":528195815,"identity":"07282304-3293-4ca0-93f7-dc73d3144c17","order_by":0,"name":"Maximus Powers","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABAklEQVRIie3PMWrDMBSAYYlAvQS8Sg34DA6CuKZDrhKR2Xsm84xBHuruOYYmt6ONoV0UvCpoCr1ASqZAhsaFDAULZ+ygb3pI/OgJIcf5z3w0AYTqfuwHRMYTCviW4DuTsP6b2EVeOf/aXKol+2yy00alQeQ14nH//oT84mU1lMSlYvOdMLxSPKdKtywuuZgliiCidnJwH51UNAOzWmgMFI41l3WfCIJCktiStzNczJJtcX6GY8pldxhNKgwPBkuCBQU94VKPvBKX3Ylm179sFRcxqJaF+pA/X5Op7S+RJ/h3v5hftO0ePtIg7NaNSUQa+MXr8GJDh7+mtgt74jiO49z8AFKjbYl8iZ9rAAAAAElFTkSuQmCC","orcid":"","institution":"Clarkson University","correspondingAuthor":true,"prefix":"","firstName":"Maximus","middleName":"","lastName":"Powers","suffix":""},{"id":528195816,"identity":"56a10c59-dfbb-492b-8b02-292ff469098f","order_by":1,"name":"Shaina Raza","email":"","orcid":"","institution":"Toronto Metropolitan University","correspondingAuthor":false,"prefix":"","firstName":"Shaina","middleName":"","lastName":"Raza","suffix":""},{"id":528195817,"identity":"8fca43a2-31d5-485b-b3f8-7903691b6fba","order_by":2,"name":"Alex Chang","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Alex","middleName":"","lastName":"Chang","suffix":""},{"id":528195818,"identity":"19fe86b0-8eee-4663-b55a-b2717801af2e","order_by":3,"name":"Rehana Riaz","email":"","orcid":"","institution":"","correspondingAuthor":false,"prefix":"","firstName":"Rehana","middleName":"","lastName":"Riaz","suffix":""},{"id":528195819,"identity":"61f34ce1-3119-4f0e-88a3-c2200da4af8f","order_by":4,"name":"Umang Mavani","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Umang","middleName":"","lastName":"Mavani","suffix":""},{"id":528195820,"identity":"107daa9d-801f-4ac8-a9ff-e28c9a7d7030","order_by":5,"name":"Harshitha Reddy Jonala","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Harshitha","middleName":"Reddy","lastName":"Jonala","suffix":""},{"id":528195821,"identity":"aaa8e894-928c-4a41-9941-0157360c82e8","order_by":6,"name":"Ansh Tiwari","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Ansh","middleName":"","lastName":"Tiwari","suffix":""},{"id":528195822,"identity":"64a6d000-bf6f-4489-b523-aa7a888e8e4b","order_by":7,"name":"Hua Wei","email":"","orcid":"","institution":"Arizona State University","correspondingAuthor":false,"prefix":"","firstName":"Hua","middleName":"","lastName":"Wei","suffix":""}],"badges":[],"createdAt":"2025-09-15 20:08:24","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-7623811/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-7623811/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":93602495,"identity":"2d0c2489-01b2-4d6b-adb3-b03ce50de797","added_by":"auto","created_at":"2025-10-15 14:50:23","extension":"json","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":9515,"visible":true,"origin":"","legend":"","description":"","filename":"f352233c12fe4ff2a3d21f1b5e3b9056.json","url":"https://assets-eu.researchsquare.com/files/rs-7623811/v1/9dbe5a0e2037853839f1e21f.json"},{"id":93602536,"identity":"4e4dd0f0-73a1-4a98-be30-cf0c5b30b688","added_by":"auto","created_at":"2025-10-15 14:50:43","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":1022123,"visible":true,"origin":"","legend":"","description":"","filename":"GUSNet.pdf","url":"https://assets-eu.researchsquare.com/files/rs-7623811/v1_covered_37017441-bf7b-4c52-828a-37f7f3cdfeb0.pdf"},{"id":93602494,"identity":"470e88dc-3207-4c50-a485-1d9015035860","added_by":"auto","created_at":"2025-10-15 14:50:21","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":6886,"visible":true,"origin":"","legend":"","description":"","filename":"AuthorshipContributionStatement.docx","url":"https://assets-eu.researchsquare.com/files/rs-7623811/v1/5e651816da086d26a906c7d0.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Responsible AI in NLP: GUS-Net Span-Level Bias Detection Dataset and Benchmark for Generalizations, Unfairness, and Stereotypes","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"knowledge-and-information-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"kais","sideBox":"Learn more about [Knowledge and Information Systems](http://link.springer.com/journal/10115)","snPcode":"10115","submissionUrl":"https://submission.nature.com/new-submission/10115/3","title":"Knowledge and Information Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Bias detection, Token-level tagging, BIO, Social bias, Responsible AI, Dataset, NLP","lastPublishedDoi":"10.21203/rs.3.rs-7623811/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-7623811/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"Representational harms in language technologies often occur in short spans within otherwise neutral text, where phrases may simultaneously convey generalizations, unfairness, or stereotypes. Framing bias detection as sentence-level classification obscures which words carry bias and what type is present, limiting both auditability and targeted mitigation. We introduce the GUS-Net Framework, comprising the GUS dataset and a multi-label token-level detector for span-level analysis of social bias. The GUS dataset contains 3,739 unique snippets across multiple domains, with over 69,000 token-level annotations. Each token is labeled using BIO tags (Begin, Inside, Outside) for three pathways of representational harm: Generalizations, Unfairness, and Stereotypes. To ensure reliable data annotation, we employ an automated multi-agent pipeline that proposes candidate spans which are subsequently verified and corrected by human experts. We formulate bias detection as multi-label token-level classification and benchmark both encoder-based models (e.g., BERT family variants) and decoder-based large language models (LLMs). Our evaluations cover token-level identification and span-level entity recognition on our test set, and out-of-distribution generalization. Empirical results show that encoder-based models consistently outperform decoder-based baselines on nuanced and overlapping spans while being more computationally efficient. The framework delivers interpretable, fine-grained diagnostics that enable systematic auditing and mitigation of representational harms in real-world NLP systems.","manuscriptTitle":"Responsible AI in NLP: GUS-Net Span-Level Bias Detection Dataset and Benchmark for Generalizations, Unfairness, and Stereotypes","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-10-15 14:47:18","doi":"10.21203/rs.3.rs-7623811/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-05-11T07:49:02+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"196710071451040115900461292601149966783","date":"2026-04-24T01:01:07+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-30T08:23:04+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"45136885678286234373541103179347691582","date":"2026-03-18T03:14:01+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"93349290022456979349216210053130765194","date":"2025-10-04T01:08:20+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2025-10-01T22:45:15+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2025-09-25T12:38:38+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2025-09-16T11:13:28+00:00","index":"","fulltext":""},{"type":"submitted","content":"Knowledge and Information Systems","date":"2025-09-15T20:06:32+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"knowledge-and-information-systems","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"kais","sideBox":"Learn more about [Knowledge and Information Systems](http://link.springer.com/journal/10115)","snPcode":"10115","submissionUrl":"https://submission.nature.com/new-submission/10115/3","title":"Knowledge and Information Systems","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"em","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"38f30077-9023-4414-ba7c-dd58ff28fb40","owner":[],"postedDate":"October 15th, 2025","published":true,"recentEditorialEvents":[{"type":"editorInvitedReview","content":"","date":"2026-05-11T07:49:02+00:00","index":82,"fulltext":""}],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2025-10-15T14:47:18+00:00","versionOfRecord":[],"versionCreatedAt":"2025-10-15 14:47:18","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-7623811","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-7623811","identity":"rs-7623811","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}