{"id":"https://openalex.org/W7138858076","doi":"https://doi.org/10.1609/aaai.v40i40.40690","title":"Finding the Translation Switch: Discovering and Exploiting the Task-Initiation Features in LLMs","display_name":"Finding the Translation Switch: Discovering and Exploiting the Task-Initiation Features in LLMs","publication_year":2026,"publication_date":"2026-03-14","ids":{"openalex":"https://openalex.org/W7138858076","doi":"https://doi.org/10.1609/aaai.v40i40.40690"},"language":null,"primary_location":{"id":"doi:10.1609/aaai.v40i40.40690","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i40.40690","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40690/44651","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40690/44651","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101626081","display_name":"Xudong Wu","orcid":"https://orcid.org/0000-0002-9231-2401"},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":true,"raw_author_name":"Xinwei Wu","raw_affiliation_strings":["Tianjin University"],"affiliations":[{"raw_affiliation_string":"Tianjin University","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130048743","display_name":"Heng Liu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Heng Liu","raw_affiliation_strings":["Alibaba International Digital Commerce"],"affiliations":[{"raw_affiliation_string":"Alibaba International Digital Commerce","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129944489","display_name":"Xiaohu Zhao","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xiaohu Zhao","raw_affiliation_strings":["Alibaba International Digital Commerce"],"affiliations":[{"raw_affiliation_string":"Alibaba International Digital Commerce","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129772678","display_name":"Yuqi Ren","orcid":null},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Yuqi Ren","raw_affiliation_strings":["Tianjin University"],"affiliations":[{"raw_affiliation_string":"Tianjin University","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129839173","display_name":"Linlong Xu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Linlong Xu","raw_affiliation_strings":["Alibaba International Digital Commerce"],"affiliations":[{"raw_affiliation_string":"Alibaba International Digital Commerce","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5129939424","display_name":"Longyue Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Longyue Wang","raw_affiliation_strings":["Alibaba International Digital Commerce"],"affiliations":[{"raw_affiliation_string":"Alibaba International Digital Commerce","institution_ids":[]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5106275949","display_name":"Deyi Xiong","orcid":null},"institutions":[{"id":"https://openalex.org/I162868743","display_name":"Tianjin University","ror":"https://ror.org/012tb2g32","country_code":"CN","type":"education","lineage":["https://openalex.org/I162868743"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Deyi Xiong","raw_affiliation_strings":["Tianjin University"],"affiliations":[{"raw_affiliation_string":"Tianjin University","institution_ids":["https://openalex.org/I162868743"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5130023045","display_name":"Weihua Luo","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Weihua Luo","raw_affiliation_strings":["Alibaba International Digital Commerce"],"affiliations":[{"raw_affiliation_string":"Alibaba International Digital Commerce","institution_ids":[]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5130088883","display_name":"Kaifu Zhang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kaifu Zhang","raw_affiliation_strings":["Alibaba International Digital Commerce"],"affiliations":[{"raw_affiliation_string":"Alibaba International Digital Commerce","institution_ids":[]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":9,"corresponding_author_ids":["https://openalex.org/A5101626081"],"corresponding_institution_ids":["https://openalex.org/I162868743"],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.73308471,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":"40","issue":"40","first_page":"33971","last_page":"33979"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.2883000075817108,"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/T10028","display_name":"Topic Modeling","score":0.2883000075817108,"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":0.1826000064611435,"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/T11636","display_name":"Artificial Intelligence in Healthcare and Education","score":0.08299999684095383,"subfield":{"id":"https://openalex.org/subfields/2718","display_name":"Health Informatics"},"field":{"id":"https://openalex.org/fields/27","display_name":"Medicine"},"domain":{"id":"https://openalex.org/domains/4","display_name":"Health Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/leverage","display_name":"Leverage (statistics)","score":0.6751000285148621},{"id":"https://openalex.org/keywords/translation","display_name":"Translation (biology)","score":0.6327000260353088},{"id":"https://openalex.org/keywords/mechanism","display_name":"Mechanism (biology)","score":0.5242999792098999},{"id":"https://openalex.org/keywords/component","display_name":"Component (thermodynamics)","score":0.45739999413490295},{"id":"https://openalex.org/keywords/blueprint","display_name":"Blueprint","score":0.42010000348091125},{"id":"https://openalex.org/keywords/boosting","display_name":"Boosting (machine learning)","score":0.3919999897480011}],"concepts":[{"id":"https://openalex.org/C153083717","wikidata":"https://www.wikidata.org/wiki/Q6535263","display_name":"Leverage (statistics)","level":2,"score":0.6751000285148621},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6643999814987183},{"id":"https://openalex.org/C149364088","wikidata":"https://www.wikidata.org/wiki/Q185917","display_name":"Translation (biology)","level":4,"score":0.6327000260353088},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5292999744415283},{"id":"https://openalex.org/C89611455","wikidata":"https://www.wikidata.org/wiki/Q6804646","display_name":"Mechanism (biology)","level":2,"score":0.5242999792098999},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.45739999413490295},{"id":"https://openalex.org/C155911762","wikidata":"https://www.wikidata.org/wiki/Q422321","display_name":"Blueprint","level":2,"score":0.42010000348091125},{"id":"https://openalex.org/C46686674","wikidata":"https://www.wikidata.org/wiki/Q466303","display_name":"Boosting (machine learning)","level":2,"score":0.3919999897480011},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.376800000667572},{"id":"https://openalex.org/C2164484","wikidata":"https://www.wikidata.org/wiki/Q5170150","display_name":"Core (optical fiber)","level":2,"score":0.3734000027179718},{"id":"https://openalex.org/C188198153","wikidata":"https://www.wikidata.org/wiki/Q1613840","display_name":"Limiting","level":2,"score":0.3628000020980835},{"id":"https://openalex.org/C177264268","wikidata":"https://www.wikidata.org/wiki/Q1514741","display_name":"Set (abstract data type)","level":2,"score":0.36070001125335693},{"id":"https://openalex.org/C203005215","wikidata":"https://www.wikidata.org/wiki/Q79798","display_name":"Machine translation","level":2,"score":0.3301999866962433},{"id":"https://openalex.org/C63479239","wikidata":"https://www.wikidata.org/wiki/Q7353546","display_name":"Robustness (evolution)","level":3,"score":0.3027999997138977},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2906000018119812}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.1609/aaai.v40i40.40690","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i40.40690","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40690/44651","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"}],"best_oa_location":{"id":"doi:10.1609/aaai.v40i40.40690","is_oa":true,"landing_page_url":"https://doi.org/10.1609/aaai.v40i40.40690","pdf_url":"https://ojs.aaai.org/index.php/AAAI/article/download/40690/44651","source":{"id":"https://openalex.org/S4210191458","display_name":"Proceedings of the AAAI Conference on Artificial Intelligence","issn_l":"2159-5399","issn":["2159-5399","2374-3468"],"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/P4310320058","host_organization_name":"Association for the Advancement of Artificial Intelligence","host_organization_lineage":["https://openalex.org/P4310320058"],"host_organization_lineage_names":["Association for the Advancement of Artificial Intelligence"],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the AAAI Conference on Artificial Intelligence","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320335777","display_name":"National Key Research and Development Program of China","ror":null}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7138858076.pdf","grobid_xml":"https://content.openalex.org/works/W7138858076.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"frequently":[4,49],"exhibit":[5],"strong":[6],"translation":[7,52,108,144,182],"abilities,":[8],"even":[9],"without":[10],"task-specific":[11,40],"fine-tuning.":[12,129],"However,":[13],"the":[14,85,105,143,169,181],"internal":[15,193],"mechanisms":[16,162],"governing":[17],"this":[18,26,117,149],"innate":[19,107],"capability":[20],"remain":[21],"largely":[22],"opaque.":[23],"To":[24],"demystify":[25],"process,":[27],"we":[28,115,131,159],"leverage":[29,116],"Sparse":[30],"Autoencoders":[31],"(SAEs)":[32],"and":[33,54,95,155,200],"introduce":[34],"a":[35,62,70,101,122,177,189],"novel":[36],"framework":[37,67],"for":[38,58,127,191],"identifying":[39],"features.":[41,76,146],"Our":[42,172],"method":[43],"first":[44],"recalls":[45],"features":[46,83],"that":[47,80,138],"are":[48,163],"co-activated":[50],"on":[51,134],"inputs":[53],"then":[55],"filters":[56],"them":[57,92],"functional":[59],"coherence":[60],"using":[61,192],"PCA-based":[63],"consistency":[64],"metric.":[65],"This":[66],"successfully":[68],"isolates":[69],"small":[71],"set":[72],"of":[73,104,168,180],"\"translation":[74],"initiation\"":[75],"Causal":[77],"interventions":[78],"demonstrate":[79],"amplifying":[81],"these":[82,161],"steers":[84],"model":[86,194],"towards":[87],"correct":[88],"translation,":[89],"while":[90],"ablating":[91],"induces":[93],"hallucinations":[94],"off-task":[96],"outputs,":[97],"confirming":[98],"they":[99],"represent":[100],"core":[102,178],"component":[103,179],"model's":[106],"competency.":[109],"Moving":[110],"from":[111],"analysis":[112],"to":[113,120,140,165,196],"application,":[114],"mechanistic":[118],"insight":[119],"propose":[121],"new":[123],"data":[124,153],"selection":[125],"strategy":[126],"efficient":[128,201],"Specifically,":[130],"prioritize":[132],"training":[133],"\"mechanistically":[135],"hard\"":[136],"samples\u2014those":[137],"fail":[139],"naturally":[141],"activate":[142],"initiation":[145],"Experiments":[147],"show":[148],"approach":[150],"significantly":[151],"improves":[152],"efficiency":[154],"suppresses":[156],"hallucinations.":[157],"Furthermore,":[158],"find":[160],"transferable":[164],"larger":[166],"models":[167],"same":[170],"family.":[171],"work":[173],"not":[174],"only":[175],"decodes":[176],"mechanism":[183,195],"in":[184],"LLMs":[185],"but":[186],"also":[187],"provides":[188],"blueprint":[190],"create":[197],"more":[198],"robust":[199],"models.":[202]},"counts_by_year":[],"updated_date":"2026-04-03T22:45:19.894376","created_date":"2026-03-20T00:00:00"}
