[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-separating-grammar-from-bias-in-language-models":10,"sections":34},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":24,"tags":25,"sources":29,"feedback":33,"feedback_at":22,"cost_usd":33,"total_tokens":33},2767,"separating-grammar-from-bias-in-language-models","Separating Grammar From Bias in Language Models","A new research framework tries to pull apart grammatical gender and social bias in contextual embeddings, a problem older debiasing tools have largely ignored.","Researchers have built a framework to disentangle grammatical gender from semantic bias in contextual language model embeddings, targeting a blind spot that existing debiasing tools have left open.\n\nMost gender debiasing work in NLP has focused on static word embeddings — fixed vector representations that do not change based on surrounding text. Contextual embeddings, the kind produced by modern language models, shift meaning based on context, which makes them harder to audit. The paper targets gendered languages like Spanish, where a noun's grammatical gender is a syntactic fact — a table is feminine, a book is masculine — but models can conflate that with social assumptions about people. The researchers built balanced datasets of inanimate nouns using controlled templates and natural Wikipedia text, then tested three direction estimators: a centroid method, a Support Vector Machine, and Linear Discriminant Analysis.\n\nThe distinction matters because debiasing a model without separating these two signals risks one of two failure modes: scrubbing grammatical gender that the language actually needs, or leaving social bias untouched while thinking the job is done. Having a dual-objective metric that scores both suppression of grammatical leakage and preservation of meaningful gender distinctions for occupation terms is a more honest test than most prior work applies.\n\nThe centroid estimator — the simplest of the three — outperformed the more complex discriminative approaches, and unweighted controlled contexts produced the cleanest grammatical gender direction. That result is a small reminder that elaborate machinery does not always win; it also means the finding is easier to reproduce and build on.","[\"ai\",\"nlp\",\"bias\",\"research\"]","2026-06-30T04:00:00.000Z","2026-06-30T12:22:05.813Z","2026-06-30T12:22:08.784Z","published",null,[],"ai",[24,26,27,28],"nlp","bias","research",[30],{"name":31,"url":32},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2606.30152",0,{"sections":35},[36,40,45,50,55,60,65,70,75,80,85,89,94,99],{"name":37,"slug":24,"count":38,"latest_published_at":39},"AI",2590,"2026-07-16T04:00:00.000Z",{"name":41,"slug":42,"count":43,"latest_published_at":44},"Security","security",294,"2026-07-15T19:59:48.000Z",{"name":46,"slug":47,"count":48,"latest_published_at":49},"Deals","deals",179,"2026-06-29T20:02:07.000Z",{"name":51,"slug":52,"count":53,"latest_published_at":54},"Policy","policy",158,"2026-07-16T00:02:48.000Z",{"name":56,"slug":57,"count":58,"latest_published_at":59},"Hardware","hardware",122,"2026-07-14T19:46:26.000Z",{"name":61,"slug":62,"count":63,"latest_published_at":64},"Consumer Tech","consumer-tech",93,"2026-07-13T13:20:48.000Z",{"name":66,"slug":67,"count":68,"latest_published_at":69},"Software","software",70,"2026-07-13T19:52:25.000Z",{"name":71,"slug":72,"count":73,"latest_published_at":74},"Science","science",66,"2026-07-10T10:29:37.000Z",{"name":76,"slug":77,"count":78,"latest_published_at":79},"Dev Tools","dev-tools",59,"2026-07-07T04:00:00.000Z",{"name":81,"slug":82,"count":83,"latest_published_at":84},"Gaming","gaming",41,"2026-07-09T04:00:00.000Z",{"name":86,"slug":87,"count":83,"latest_published_at":88},"Startups","startups","2026-06-29T20:55:50.000Z",{"name":90,"slug":91,"count":92,"latest_published_at":93},"General","general",29,"2026-07-10T22:28:58.000Z",{"name":95,"slug":96,"count":97,"latest_published_at":98},"Reviews","reviews",20,"2026-06-24T12:00:01.000Z",{"name":100,"slug":101,"count":102,"latest_published_at":103},"How-To","how-to",6,"2026-06-16T09:00:00.000Z"]