{"id":"https://openalex.org/W7224499501","doi":"https://doi.org/10.18653/2026.emnlp-main.1199","title":"Does Alignment Debias or Just Suppress? Evaluating Stereotypical Bias Across Base-Instruct Pairs","display_name":"Does Alignment Debias or Just Suppress? Evaluating Stereotypical Bias Across Base-Instruct Pairs","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7224499501","doi":"https://doi.org/10.18653/2026.emnlp-main.1199"},"language":"en","primary_location":{"id":"doi:10.18653/2026.emnlp-main.1199","doi":"https://doi.org/10.18653/2026.emnlp-main.1199","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.1199","pdf_url":"https://aclanthology.org/2026.emnlp-main.1199.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2026.emnlp-main.1199.pdf","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5160068343","display_name":"Sigurd Kornelius Havstein","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I124055696","display_name":"University of Copenhagen","ror":"https://ror.org/035b05819","country_code":"DK","type":"education","lineage":["https://openalex.org/I124055696"]}],"countries":["DK"],"is_corresponding":false,"raw_author_name":"Sigurd Kornelius Havstein","raw_affiliation_strings":["Department of Computer Science University of Copenhagen Copenhagen , Denmark"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of Copenhagen Copenhagen , Denmark","institution_ids":["https://openalex.org/I124055696"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5139768385","display_name":"Anders S\u00f8gaard","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I124055696","display_name":"University of Copenhagen","ror":"https://ror.org/035b05819","country_code":"DK","type":"education","lineage":["https://openalex.org/I124055696"]}],"countries":["DK"],"is_corresponding":false,"raw_author_name":"Anders S\u00f8gaard","raw_affiliation_strings":["Department of Computer Science University of Copenhagen Copenhagen , Denmark"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of Copenhagen Copenhagen , Denmark","institution_ids":["https://openalex.org/I124055696"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I124055696"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.92141839,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"27446","last_page":"27460"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.3684000074863434,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10883","display_name":"Ethics and Social Impacts of AI","score":0.3684000074863434,"subfield":{"id":"https://openalex.org/subfields/3311","display_name":"Safety Research"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.298799991607666,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}},{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.1354999989271164,"subfield":{"id":"https://openalex.org/subfields/1702","display_name":"Artificial Intelligence"},"field":{"id":"https://openalex.org/fields/17","display_name":"Computer Science"},"domain":{"id":"https://openalex.org/domains/3","display_name":"Physical Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/stereotypical-bias","display_name":"stereotypical bias","score":0.7760000228881836},{"id":"https://openalex.org/keywords/instruction-tuning","display_name":"instruction tuning","score":0.527999997138977},{"id":"https://openalex.org/keywords/language-model-alignment","display_name":"language model alignment","score":0.503000020980835},{"id":"https://openalex.org/keywords/bias-suppression","display_name":"bias suppression","score":0.5460000038146973}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4415999948978424},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.43149998784065247},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3772999942302704},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.37389999628067017},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.34470000863075256},{"id":"https://openalex.org/C99498987","wikidata":"https://www.wikidata.org/wiki/Q2210247","display_name":"Noise (video)","level":3,"score":0.3206999897956848},{"id":"https://openalex.org/C26760741","wikidata":"https://www.wikidata.org/wiki/Q160402","display_name":"Perception","level":2,"score":0.3001999855041504},{"id":"https://openalex.org/C116834253","wikidata":"https://www.wikidata.org/wiki/Q2039217","display_name":"Identification (biology)","level":2,"score":0.2809000015258789},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.27059999108314514},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2700999975204468},{"id":"https://openalex.org/C12713177","wikidata":"https://www.wikidata.org/wiki/Q1900281","display_name":"Perspective (graphical)","level":2,"score":0.26100000739097595}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/2026.emnlp-main.1199","doi":"https://doi.org/10.18653/2026.emnlp-main.1199","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.1199","pdf_url":"https://aclanthology.org/2026.emnlp-main.1199.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/2026.emnlp-main.1199","doi":"https://doi.org/10.18653/2026.emnlp-main.1199","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.emnlp-main.1199","pdf_url":"https://aclanthology.org/2026.emnlp-main.1199.pdf","source":{"id":"https://openalex.org/S4363608991","display_name":"Conference on Empirical Methods in Natural Language Processing (EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224499501.pdf","grobid_xml":"https://content.openalex.org/works/W7224499501.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4401042308","https://openalex.org/W3156714224","https://openalex.org/W3019416653","https://openalex.org/W4385570906","https://openalex.org/W3168901128","https://openalex.org/W7224532337","https://openalex.org/W7145639829","https://openalex.org/W4412944562","https://openalex.org/W7224293951","https://openalex.org/W7131779109"],"abstract_inverted_index":{"Warning:":[0],"This":[1],"paper":[2],"contains":[3],"offensive":[4],"stereotypes":[5],"about":[6,65],"gender,":[7],"race,":[8],"and":[9,45,72,133,176],"religion.Alignment":[10],"has":[11],"been":[12],"widely":[13],"reported":[14],"to":[15,156],"reduce":[16],"stereotypical":[17],"bias":[18,47,141,145,155,200],"in":[19,102],"large":[20],"language":[21],"models,":[22],"but":[23],"it":[24,83,92,163],"is":[25,58,146],"unclear":[26],"whether":[27],"these":[28,190],"reductions":[29],"reflect":[30],"debiasing":[31],"or":[32],"suppression":[33],"at":[34,106],"the":[35,55,67,107,111,148],"output":[36],"level.Across":[37],"27":[38,80,90],"base-instruct":[39],"pairs":[40],"from":[41,117],"four":[42,46],"open-weight":[43],"families":[44],"benchmarks,":[48],"three":[49],"patterns":[50],"provide":[51],"substantial":[52],"evidence":[53],"that":[54],"measured":[56],"reduction":[57],"largely":[59],"suppression.(i)":[60],"The":[61],"chat":[62],"template":[63,114],"drives":[64],"half":[66],"alignment":[68,196],"effect":[69],"on":[70,75,161],"CrowS-Pairs":[71,154],"essentially":[73],"all":[74],"StereoSet,":[76],"where":[77],"18":[78],"of":[79,89,110],"improve":[81],"without":[82,167],"(p":[84,93],"=":[85,100],"0.061)":[86],"versus":[87],"26":[88],"with":[91,113],"<":[94],"0.001).Per-pair":[95],"effects":[96],"remain":[97,180],"below":[98],"d":[99],"0.2":[101],"both":[103],"conditions.IAT":[104],"lies":[105],"weight-level":[108],"end":[109],"spectrum,":[112],"contribution":[115],"indistinguishable":[116],"zero.(ii)":[118],"Decomposing":[119],"BBQ's":[120],"ambiguous-context":[121],"score":[122],"shows":[123],"instruct":[124,177],"models":[125],"defer":[126],"rather":[127],"than":[128],"debias":[129],"(28%":[130],"\u2192":[131,143],"53%)":[132],"are":[134],"more":[135],"biased":[136],"when":[137],"they":[138],"answer":[139],"(conditional":[140],"0.53":[142],"0.59).Suppressed":[144],"recoverable:":[147],"most":[149],"aggressive":[150],"adversarial":[151],"framing":[152],"restores":[153],"near":[157],"base":[158,175],"levels,":[159],"while":[160],"BBQ":[162],"partly":[164],"reverses":[165],"deferral":[166],"reactivating":[168],"conditional":[169],"bias.(iii)":[170],"Across":[171],"eight":[172],"probed":[173],"pairs,":[174],"gender":[178],"directions":[179],"geometrically":[181],"aligned":[182],"(median":[183],"cosine":[184],"0.86":[185],"vs.":[186],"0.006":[187],"randompair":[188],"baseline).Together,":[189],"findings":[191],"suggest":[192],"benchmark":[193],"improvements":[194],"after":[195],"may":[197],"overstate":[198],"genuine":[199],"reduction.":[201]},"counts_by_year":[],"updated_date":"2026-10-31T12:09:11.096373","created_date":"2026-10-30T00:00:00"}