{"id":"https://openalex.org/W4412945014","doi":"https://doi.org/10.18653/v1/2025.acl-long.1007","title":"Multi-Attribute Steering of Language Models via Targeted Intervention","display_name":"Multi-Attribute Steering of Language Models via Targeted Intervention","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412945014","doi":"https://doi.org/10.18653/v1/2025.acl-long.1007"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.1007","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1007","pdf_url":"https://aclanthology.org/2025.acl-long.1007.pdf","source":null,"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 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.acl-long.1007.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101928156","display_name":"Duy Lap Nguyen","orcid":"https://orcid.org/0009-0001-4239-9174"},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]},{"id":"https://openalex.org/I1333535994","display_name":"University of North Carolina Health Care","ror":"https://ror.org/00qz24g20","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I1333535994"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Duy Nguyen","raw_affiliation_strings":["UNC Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UNC Chapel Hill","institution_ids":["https://openalex.org/I114027177","https://openalex.org/I1333535994"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5039920828","display_name":"Archiki Prasad","orcid":null},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]},{"id":"https://openalex.org/I1333535994","display_name":"University of North Carolina Health Care","ror":"https://ror.org/00qz24g20","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I1333535994"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Archiki Prasad","raw_affiliation_strings":["UNC Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UNC Chapel Hill","institution_ids":["https://openalex.org/I114027177","https://openalex.org/I1333535994"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5038276570","display_name":"Elias Stengel-Eskin","orcid":"https://orcid.org/0000-0002-6689-505X"},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]},{"id":"https://openalex.org/I1333535994","display_name":"University of North Carolina Health Care","ror":"https://ror.org/00qz24g20","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I1333535994"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Elias Stengel-Eskin","raw_affiliation_strings":["UNC Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UNC Chapel Hill","institution_ids":["https://openalex.org/I114027177","https://openalex.org/I1333535994"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5089199686","display_name":"Mohit Bansal","orcid":"https://orcid.org/0000-0001-5858-9944"},"institutions":[{"id":"https://openalex.org/I114027177","display_name":"University of North Carolina at Chapel Hill","ror":"https://ror.org/0130frc33","country_code":"US","type":"education","lineage":["https://openalex.org/I114027177"]},{"id":"https://openalex.org/I1333535994","display_name":"University of North Carolina Health Care","ror":"https://ror.org/00qz24g20","country_code":"US","type":"healthcare","lineage":["https://openalex.org/I1333535994"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Mohit Bansal","raw_affiliation_strings":["UNC Chapel Hill"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"UNC Chapel Hill","institution_ids":["https://openalex.org/I114027177","https://openalex.org/I1333535994"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"20619","last_page":"20634"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9241999983787537,"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"}},"topics":[{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9241999983787537,"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/T10028","display_name":"Topic Modeling","score":0.9059000015258789,"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/computer-science","display_name":"Computer science","score":0.638752281665802},{"id":"https://openalex.org/keywords/intervention","display_name":"Intervention (counseling)","score":0.4577793478965759},{"id":"https://openalex.org/keywords/human\u2013computer-interaction","display_name":"Human\u2013computer interaction","score":0.325296550989151},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.1710214614868164}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.638752281665802},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.4577793478965759},{"id":"https://openalex.org/C107457646","wikidata":"https://www.wikidata.org/wiki/Q207434","display_name":"Human\u2013computer interaction","level":1,"score":0.325296550989151},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.1710214614868164},{"id":"https://openalex.org/C118552586","wikidata":"https://www.wikidata.org/wiki/Q7867","display_name":"Psychiatry","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.1007","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1007","pdf_url":"https://aclanthology.org/2025.acl-long.1007.pdf","source":null,"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 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.acl-long.1007","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1007","pdf_url":"https://aclanthology.org/2025.acl-long.1007.pdf","source":null,"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 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G5643777851","display_name":null,"funder_award_id":"DRL-2112635","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G572880119","display_name":"CAREER: Semantic Multi-Task Learning for Generalizable and Interpretable Language Generation","funder_award_id":"1846185","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G580734012","display_name":"AI Institute for Engaged Learning","funder_award_id":"2112635","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"},{"id":"https://openalex.org/G6671297155","display_name":null,"funder_award_id":"CAREER","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320332180","display_name":"Defense Advanced Research Projects Agency","ror":"https://ror.org/02caytj08"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412945014.pdf","grobid_xml":"https://content.openalex.org/works/W4412945014.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2390279801","https://openalex.org/W4391913857","https://openalex.org/W2358668433","https://openalex.org/W4396701345","https://openalex.org/W2376932109","https://openalex.org/W2001405890","https://openalex.org/W4396696052"],"abstract_inverted_index":{"Inference-time":[0],"intervention":[1,70],"(ITI)":[2],"has":[3],"emerged":[4],"as":[5,47],"a":[6,17,62],"promising":[7],"method":[8],"for":[9,67,103],"steering":[10,64,75],"large":[11],"language":[12],"model":[13],"(LLM)":[14],"behavior":[15],"in":[16,112],"particular":[18],"direction":[19],"(e.g.,":[20,156],"improving":[21],"helpfulness)":[22],"by":[23],"intervening":[24],"on":[25,117,132],"token":[26],"representations":[27,86],"without":[28],"costly":[29],"updates":[30],"to":[31,39,41,91],"the":[32,83,169],"LLM's":[33],"parameters.However,":[34],"existing":[35,146],"ITI":[36,147,171],"approaches":[37,151],"fail":[38],"scale":[40],"multiattribute":[42],"settings":[43],"with":[44],"conflicts,":[45],"such":[46],"enhancing":[48],"helpfulness":[49],"while":[50,96],"also":[51],"reducing":[52,107],"toxicity.To":[53],"address":[54],"this,":[55],"we":[56,123,136],"introduce":[57],"Multi-Attribute":[58],"Targeted":[59],"Steering":[60],"(MAT-STEER),":[61],"novel":[63],"framework":[65],"designed":[66],"selective":[68],"token-level":[69],"across":[71,152,161],"multiple":[72],"attributes.MAT-STEER":[73],"learns":[74],"vectors":[76,102],"using":[77],"an":[78],"alignment":[79],"objective":[80],"that":[81],"shifts":[82],"model's":[84],"internal":[85],"of":[87,93],"undesirable":[88],"outputs":[89],"closer":[90],"those":[92],"desirable":[94],"ones":[95],"enforcing":[97],"sparsity":[98],"and":[99,129,143,148,164],"orthogonality":[100],"among":[101],"different":[104],"attributes,":[105],"thereby":[106],"inter-attribute":[108],"conflicts.We":[109],"evaluate":[110],"MAT-STEER":[111],"two":[113],"distinct":[114],"settings:":[115],"(i)":[116],"question":[118],"answering":[119],"(QA)":[120],"tasks":[121,134,163],"where":[122,135],"balance":[124],"attributes":[125,139],"like":[126,140],"truthfulness,":[127],"bias,":[128],"toxicity;":[130],"(ii)":[131],"generative":[133],"simultaneously":[137],"improve":[138],"helpfulness,":[141],"correctness,":[142],"coherence.MAT-STEER":[144],"outperforms":[145],"parameter-efficient":[149],"finetuning":[150],"both":[153],"task":[154],"types":[155],"3%":[157],"average":[158],"accuracy":[159],"gain":[160],"QA":[162],"55.82%":[165],"win":[166],"rate":[167],"against":[168],"best":[170],"baseline).":[172]},"counts_by_year":[{"year":2025,"cited_by_count":1}],"updated_date":"2026-07-22T07:51:19.307946","created_date":"2025-10-10T00:00:00"}