{"id":"https://openalex.org/W7166825235","doi":"https://doi.org/10.18653/v1/2026.findings-acl.577","title":"CausalDetox: Causal Head Selection and Intervention for Language Model Detoxification","display_name":"CausalDetox: Causal Head Selection and Intervention for Language Model Detoxification","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7166825235","doi":"https://doi.org/10.18653/v1/2026.findings-acl.577"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2026.findings-acl.577","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.577","pdf_url":"https://aclanthology.org/2026.findings-acl.577.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":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2026.findings-acl.577.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5123394653","display_name":"Yian Wang","orcid":null},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yian Wang","raw_affiliation_strings":["Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5127863965","display_name":"Yuen Chen","orcid":null},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yuen Chen","raw_affiliation_strings":["Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139827300","display_name":"Agam Goyal","orcid":null},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Agam Goyal","raw_affiliation_strings":["Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801","institution_ids":["https://openalex.org/I157725225"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5018532037","display_name":"Hari Sundaram","orcid":"https://orcid.org/0000-0003-3315-6055"},"institutions":[{"id":"https://openalex.org/I157725225","display_name":"University of Illinois Urbana-Champaign","ror":"https://ror.org/047426m28","country_code":"US","type":"education","lineage":["https://openalex.org/I157725225"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Hari Sundaram","raw_affiliation_strings":["Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Department of Computer Science University of Illinois , Urbana-Champaign Champaign , IL 61801","institution_ids":["https://openalex.org/I157725225"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I157725225"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.81466113,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"11893","last_page":"11914"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11303","display_name":"Bayesian Modeling and Causal Inference","score":0.1875,"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/T11303","display_name":"Bayesian Modeling and Causal Inference","score":0.1875,"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":0.18330000340938568,"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/T12026","display_name":"Explainable Artificial Intelligence (XAI)","score":0.15379999577999115,"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/detoxification","display_name":"Detoxification (alternative medicine)","score":0.5860000252723694},{"id":"https://openalex.org/keywords/intervention","display_name":"Intervention (counseling)","score":0.5734000205993652},{"id":"https://openalex.org/keywords/selection","display_name":"Selection (genetic algorithm)","score":0.5002999901771545},{"id":"https://openalex.org/keywords/head","display_name":"Head (geology)","score":0.37059998512268066},{"id":"https://openalex.org/keywords/causal-model","display_name":"Causal model","score":0.36730000376701355},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.31310001015663147}],"concepts":[{"id":"https://openalex.org/C2780497538","wikidata":"https://www.wikidata.org/wiki/Q1192006","display_name":"Detoxification (alternative medicine)","level":3,"score":0.5860000252723694},{"id":"https://openalex.org/C2780665704","wikidata":"https://www.wikidata.org/wiki/Q959298","display_name":"Intervention (counseling)","level":2,"score":0.5734000205993652},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.5031999945640564},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5002999901771545},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.46320000290870667},{"id":"https://openalex.org/C2780312720","wikidata":"https://www.wikidata.org/wiki/Q5689100","display_name":"Head (geology)","level":2,"score":0.37059998512268066},{"id":"https://openalex.org/C11671645","wikidata":"https://www.wikidata.org/wiki/Q5054567","display_name":"Causal model","level":2,"score":0.36730000376701355},{"id":"https://openalex.org/C71924100","wikidata":"https://www.wikidata.org/wiki/Q11190","display_name":"Medicine","level":0,"score":0.35199999809265137},{"id":"https://openalex.org/C138496976","wikidata":"https://www.wikidata.org/wiki/Q175002","display_name":"Developmental psychology","level":1,"score":0.33970001339912415},{"id":"https://openalex.org/C99508421","wikidata":"https://www.wikidata.org/wiki/Q2678675","display_name":"Physical medicine and rehabilitation","level":1,"score":0.33340001106262207},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.3301999866962433},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.31310001015663147},{"id":"https://openalex.org/C64357122","wikidata":"https://www.wikidata.org/wiki/Q1149766","display_name":"Causality (physics)","level":2,"score":0.28139999508857727},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.2648000121116638},{"id":"https://openalex.org/C3017944768","wikidata":"https://www.wikidata.org/wiki/Q1450463","display_name":"Poison control","level":2,"score":0.2531999945640564},{"id":"https://openalex.org/C2778117688","wikidata":"https://www.wikidata.org/wiki/Q264024","display_name":"Animal model","level":2,"score":0.2531000077724457}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2026.findings-acl.577","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.577","pdf_url":"https://aclanthology.org/2026.findings-acl.577.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":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-acl.577","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-acl.577","pdf_url":"https://aclanthology.org/2026.findings-acl.577.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":"Findings of the Association for Computational Linguistics: ACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7166825235.pdf","grobid_xml":"https://content.openalex.org/works/W7166825235.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"language":[1],"models":[2],"(LLMs)":[3],"frequently":[4],"generate":[5],"toxic":[6,41,91],"content,":[7],"posing":[8],"significant":[9],"risks":[10],"for":[11,40,62,81],"safe":[12],"deployment.Current":[13],"mitigation":[14],"strategies":[15],"often":[16],"degrade":[17],"generation":[18],"quality":[19],"or":[20],"require":[21],"costly":[22],"human":[23],"annotation.We":[24],"propose":[25],"CAUSALDETOX,":[26],"a":[27,52,96,132],"framework":[28],"that":[29,57,114],"identifies":[30],"and":[31,47,60,84,111,130],"intervenes":[32],"on":[33,108],"the":[34,43],"specific":[35],"attention":[36],"heads":[37,56],"causally":[38],"responsible":[39],"generation.Using":[42],"Probability":[44],"of":[45,55,99],"Necessity":[46],"Sufficiency":[48],"(PNS),":[49],"we":[50],"isolate":[51],"minimal":[53],"set":[54],"are":[58],"necessary":[59],"sufficient":[61],"toxicity.We":[63],"utilize":[64],"these":[65],"components":[66],"via":[67],"two":[68],"complementary":[69],"strategies:":[70],"(1)":[71],"Local":[72],"Inference-Time":[73],"Intervention,":[74],"which":[75,88],"constructs":[76],"dynamic,":[77],"input-specific":[78],"steering":[79],"vectors":[80],"context-aware":[82],"detoxification,":[83],"(2)":[85],"PNS-Guided":[86],"Fine-Tuning,":[87],"permanently":[89],"unlearns":[90],"representations.We":[92],"also":[93],"introduce":[94],"PARATOX,":[95],"novel":[97],"benchmark":[98],"aligned":[100],"toxic/non-toxic":[101],"sentence":[102],"pairs":[103],"enabling":[104],"controlled":[105],"counterfactual":[106],"evaluation.Experiments":[107],"ToxiGen,":[109],"ImplicitHate,":[110],"Pa-raDetox":[112],"show":[113],"CAUSALDETOX":[115],"achieves":[116],"up":[117],"to":[118,124],"5.34%":[119],"greater":[120],"toxicity":[121],"reduction":[122],"compared":[123],"baselines":[125],"while":[126],"preserving":[127],"linguistic":[128],"fluency,":[129],"offers":[131],"7\u00d7":[133],"speedup":[134],"in":[135],"head":[136],"selection.":[137]},"counts_by_year":[],"updated_date":"2026-07-29T14:22:42.915294","created_date":"2026-07-02T00:00:00"}