{"id":"https://openalex.org/W4403297898","doi":"https://doi.org/10.1162/tacl_a_00701","title":"Not Eliminate but Aggregate: Post-Hoc Control over Mixture-of-Experts to Address Shortcut Shifts in Natural Language Understanding","display_name":"Not Eliminate but Aggregate: Post-Hoc Control over Mixture-of-Experts to Address Shortcut Shifts in Natural Language Understanding","publication_year":2024,"publication_date":"2024-01-01","ids":{"openalex":"https://openalex.org/W4403297898","doi":"https://doi.org/10.1162/tacl_a_00701"},"language":"en","primary_location":{"id":"doi:10.1162/tacl_a_00701","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00701","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00701/2480600/tacl_a_00701.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"type":"article","indexed_in":["crossref","doaj"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00701/2480600/tacl_a_00701.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5071438625","display_name":"Ukyo Honda","orcid":"https://orcid.org/0000-0002-4894-9886"},"institutions":[{"id":"https://openalex.org/I4210089607","display_name":"CyberAgent (Japan)","ror":"https://ror.org/0060jg679","country_code":"JP","type":"company","lineage":["https://openalex.org/I4210089607"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Ukyo Honda","raw_affiliation_strings":["CyberAgent, Japan. honda_ukyo@cyberagent.co.jp"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"CyberAgent, Japan. honda_ukyo@cyberagent.co.jp","institution_ids":["https://openalex.org/I4210089607"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5079445213","display_name":"Tatsushi Oka","orcid":null},"institutions":[{"id":"https://openalex.org/I203951103","display_name":"Keio University","ror":"https://ror.org/02kn6nx58","country_code":"JP","type":"education","lineage":["https://openalex.org/I203951103"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Tatsushi Oka","raw_affiliation_strings":["Keio University, Japan. tatsushi.oka@keio.jp"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Keio University, Japan. tatsushi.oka@keio.jp","institution_ids":["https://openalex.org/I203951103"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5090744564","display_name":"Peinan Zhang","orcid":"https://orcid.org/0009-0007-6541-1029"},"institutions":[{"id":"https://openalex.org/I4210089607","display_name":"CyberAgent (Japan)","ror":"https://ror.org/0060jg679","country_code":"JP","type":"company","lineage":["https://openalex.org/I4210089607"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Peinan Zhang","raw_affiliation_strings":["CyberAgent, Japan. zhang_peinan@cyberagent.co.jp"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"CyberAgent, Japan. zhang_peinan@cyberagent.co.jp","institution_ids":["https://openalex.org/I4210089607"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5003760956","display_name":"Masato Mita","orcid":"https://orcid.org/0000-0001-6210-3716"},"institutions":[{"id":"https://openalex.org/I4210089607","display_name":"CyberAgent (Japan)","ror":"https://ror.org/0060jg679","country_code":"JP","type":"company","lineage":["https://openalex.org/I4210089607"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Masato Mita","raw_affiliation_strings":["CyberAgent, Japan. mita_masato@cyberagent.co.jp"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"CyberAgent, Japan. mita_masato@cyberagent.co.jp","institution_ids":["https://openalex.org/I4210089607"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.4647,"has_fulltext":true,"cited_by_count":2,"citation_normalized_percentile":{"value":0.71884767,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":95,"max":96},"biblio":{"volume":"12","issue":null,"first_page":"1268","last_page":"1289"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9997000098228455,"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":0.9997000098228455,"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":0.9987000226974487,"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/T11714","display_name":"Multimodal Machine Learning Applications","score":0.9873999953269958,"subfield":{"id":"https://openalex.org/subfields/1707","display_name":"Computer Vision and Pattern Recognition"},"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":0.8466732501983643},{"id":"https://openalex.org/keywords/aggregate","display_name":"Aggregate (composite)","score":0.7540217638015747},{"id":"https://openalex.org/keywords/control","display_name":"Control (management)","score":0.6202319264411926},{"id":"https://openalex.org/keywords/post-hoc","display_name":"Post hoc","score":0.5747719407081604},{"id":"https://openalex.org/keywords/natural","display_name":"Natural (archaeology)","score":0.50911945104599},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.35208582878112793},{"id":"https://openalex.org/keywords/history","display_name":"History","score":0.05342757701873779}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.8466732501983643},{"id":"https://openalex.org/C4679612","wikidata":"https://www.wikidata.org/wiki/Q866298","display_name":"Aggregate (composite)","level":2,"score":0.7540217638015747},{"id":"https://openalex.org/C2775924081","wikidata":"https://www.wikidata.org/wiki/Q55608371","display_name":"Control (management)","level":2,"score":0.6202319264411926},{"id":"https://openalex.org/C2992886853","wikidata":"https://www.wikidata.org/wiki/Q18381816","display_name":"Post hoc","level":2,"score":0.5747719407081604},{"id":"https://openalex.org/C2776608160","wikidata":"https://www.wikidata.org/wiki/Q4785462","display_name":"Natural (archaeology)","level":2,"score":0.50911945104599},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.35208582878112793},{"id":"https://openalex.org/C95457728","wikidata":"https://www.wikidata.org/wiki/Q309","display_name":"History","level":0,"score":0.05342757701873779},{"id":"https://openalex.org/C166957645","wikidata":"https://www.wikidata.org/wiki/Q23498","display_name":"Archaeology","level":1,"score":0.0},{"id":"https://openalex.org/C192562407","wikidata":"https://www.wikidata.org/wiki/Q228736","display_name":"Materials science","level":0,"score":0.0},{"id":"https://openalex.org/C159985019","wikidata":"https://www.wikidata.org/wiki/Q181790","display_name":"Composite material","level":1,"score":0.0},{"id":"https://openalex.org/C71924100","wikidata":"https://www.wikidata.org/wiki/Q11190","display_name":"Medicine","level":0,"score":0.0},{"id":"https://openalex.org/C199343813","wikidata":"https://www.wikidata.org/wiki/Q12128","display_name":"Dentistry","level":1,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1162/tacl_a_00701","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00701","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00701/2480600/tacl_a_00701.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},{"id":"pmh:oai:doaj.org/article:977850bc99a24c9ba4c0c92bd84e749c","is_oa":true,"landing_page_url":"https://doaj.org/article/977850bc99a24c9ba4c0c92bd84e749c","pdf_url":null,"source":{"id":"https://openalex.org/S4306401280","display_name":"DOAJ (DOAJ: Directory of Open Access Journals)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by-sa","license_id":"https://openalex.org/licenses/cc-by-sa","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"Transactions of the Association for Computational Linguistics, Vol 12 (2024)","raw_type":"article"}],"best_oa_location":{"id":"doi:10.1162/tacl_a_00701","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00701","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00701/2480600/tacl_a_00701.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4403297898.pdf","grobid_xml":"https://content.openalex.org/works/W4403297898.grobid-xml"},"referenced_works_count":87,"referenced_works":["https://openalex.org/W1988520084","https://openalex.org/W2056179483","https://openalex.org/W2150884987","https://openalex.org/W2229667755","https://openalex.org/W2467153441","https://openalex.org/W2604738573","https://openalex.org/W2797746357","https://openalex.org/W2913668833","https://openalex.org/W2951286828","https://openalex.org/W2962736243","https://openalex.org/W2963341956","https://openalex.org/W2963846996","https://openalex.org/W2963961878","https://openalex.org/W2964044490","https://openalex.org/W2965373594","https://openalex.org/W2970019270","https://openalex.org/W2970379526","https://openalex.org/W2970442950","https://openalex.org/W2979840691","https://openalex.org/W2984256198","https://openalex.org/W3019941393","https://openalex.org/W3034831508","https://openalex.org/W3035139434","https://openalex.org/W3100511085","https://openalex.org/W3100895823","https://openalex.org/W3102259594","https://openalex.org/W3104169042","https://openalex.org/W3105302490","https://openalex.org/W3152911627","https://openalex.org/W3155431862","https://openalex.org/W3170037207","https://openalex.org/W3175352680","https://openalex.org/W3177934633","https://openalex.org/W4205509257","https://openalex.org/W4212774754","https://openalex.org/W4225468040","https://openalex.org/W4280586943","https://openalex.org/W4283792669","https://openalex.org/W4283796213","https://openalex.org/W4287391717","https://openalex.org/W4287854873","https://openalex.org/W4288287305","https://openalex.org/W4291817780","https://openalex.org/W4296413648","https://openalex.org/W4297631950","https://openalex.org/W4302305855","https://openalex.org/W4312516176","https://openalex.org/W4385571040","https://openalex.org/W4385572739","https://openalex.org/W4385572896","https://openalex.org/W4385572918","https://openalex.org/W4385573905","https://openalex.org/W4385574074","https://openalex.org/W4385574188","https://openalex.org/W4386576728","https://openalex.org/W4394595682","https://openalex.org/W6732520560","https://openalex.org/W6735377749","https://openalex.org/W6745265922","https://openalex.org/W6763701032","https://openalex.org/W6765039553","https://openalex.org/W6765285020","https://openalex.org/W6766673545","https://openalex.org/W6768920361","https://openalex.org/W6771898459","https://openalex.org/W6771917389","https://openalex.org/W6774630179","https://openalex.org/W6780805062","https://openalex.org/W6783608825","https://openalex.org/W6784706293","https://openalex.org/W6787241537","https://openalex.org/W6788811087","https://openalex.org/W6791853581","https://openalex.org/W6794803083","https://openalex.org/W6796219925","https://openalex.org/W6803557570","https://openalex.org/W6804097997","https://openalex.org/W6804272589","https://openalex.org/W6809924826","https://openalex.org/W6811172144","https://openalex.org/W6811242705","https://openalex.org/W6840963292","https://openalex.org/W6843384352","https://openalex.org/W6845120169","https://openalex.org/W6845955646","https://openalex.org/W6846121321","https://openalex.org/W6850164823"],"related_works":["https://openalex.org/W2485872624","https://openalex.org/W2182025991","https://openalex.org/W2023578311","https://openalex.org/W4283262748","https://openalex.org/W4324133059","https://openalex.org/W1800639126","https://openalex.org/W1833397253","https://openalex.org/W2370273288","https://openalex.org/W2368441895","https://openalex.org/W3103272396"],"abstract_inverted_index":{"Abstract":[0],"Recent":[1],"models":[2,39,68],"for":[3],"natural":[4],"language":[5],"understanding":[6],"are":[7,40,54],"inclined":[8],"to":[9,42,69,114,138],"exploit":[10],"simple":[11],"patterns":[12],"in":[13,31,118],"datasets,":[14],"commonly":[15],"known":[16],"as":[17],"shortcuts.":[18,74,119],"These":[19],"shortcuts":[20],"hinge":[21],"on":[22,73],"spurious":[23],"correlations":[24],"between":[25],"labels":[26],"and":[27,135],"latent":[28,52,96],"features":[29,53],"existing":[30],"the":[32,59,71,85,107,111,115,140],"training":[33],"data.":[34],"At":[35],"inference":[36],"time,":[37],"shortcut-dependent":[38],"likely":[41],"generate":[43],"erroneous":[44],"predictions":[45,86],"under":[46],"distribution":[47,116],"shifts,":[48],"particularly":[49],"when":[50],"some":[51,127],"no":[55],"longer":[56],"correlated":[57],"with":[58],"labels.":[60],"To":[61],"avoid":[62],"this,":[63],"previous":[64],"studies":[65],"have":[66],"trained":[67],"eliminate":[70],"reliance":[72],"In":[75],"this":[76],"study,":[77],"we":[78,121],"explore":[79],"a":[80,88],"different":[81,95],"direction:":[82],"pessimistically":[83],"aggregating":[84],"of":[87],"mixture-of-experts,":[89],"assuming":[90],"each":[91],"expert":[92],"captures":[93],"relatively":[94],"features.":[97],"The":[98],"experimental":[99],"results":[100,137],"demonstrate":[101],"that":[102,123],"our":[103,124,133],"post-hoc":[104],"control":[105],"over":[106],"experts":[108],"significantly":[109],"enhances":[110],"model\u2019s":[112],"robustness":[113],"shift":[117],"Additionally,":[120],"show":[122],"approach":[125],"has":[126],"practical":[128],"advantages.":[129],"We":[130],"also":[131],"analyze":[132],"model":[134],"provide":[136],"support":[139],"assumption.1":[141]},"counts_by_year":[{"year":2025,"cited_by_count":2}],"updated_date":"2026-08-21T09:56:20.448147","created_date":"2024-10-11T00:00:00"}