{"id":"https://openalex.org/W4412889746","doi":"https://doi.org/10.18653/v1/2025.acl-long.1458","title":"Stochastic Chameleons: Irrelevant Context Hallucinations Reveal Class-Based (Mis)Generalization in LLMs","display_name":"Stochastic Chameleons: Irrelevant Context Hallucinations Reveal Class-Based (Mis)Generalization in LLMs","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412889746","doi":"https://doi.org/10.18653/v1/2025.acl-long.1458"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.1458","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1458","pdf_url":"https://aclanthology.org/2025.acl-long.1458.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":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.acl-long.1458.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5101290233","display_name":"Ziling Cheng","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ziling Cheng","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101856888","display_name":"Meng Cao","orcid":"https://orcid.org/0000-0003-2657-7142"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Meng Cao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5087734326","display_name":"Marc-Antoine Rondeau","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Marc-Antoine Rondeau","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":null,"display_name":"Jackie CK Cheung","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jackie CK Cheung","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":1.7588,"has_fulltext":true,"cited_by_count":1,"citation_normalized_percentile":{"value":0.87802391,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":{"min":95,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"30187","last_page":"30214"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11512","display_name":"Anomaly Detection Techniques and Applications","score":0.9194999933242798,"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/T11512","display_name":"Anomaly Detection Techniques and Applications","score":0.9194999933242798,"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/generalization","display_name":"Generalization","score":0.7578318119049072},{"id":"https://openalex.org/keywords/context","display_name":"Context (archaeology)","score":0.6409345865249634},{"id":"https://openalex.org/keywords/class","display_name":"Class (philosophy)","score":0.6213930249214172},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.48993271589279175},{"id":"https://openalex.org/keywords/cognitive-psychology","display_name":"Cognitive psychology","score":0.4399090111255646},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.41161638498306274},{"id":"https://openalex.org/keywords/psychology","display_name":"Psychology","score":0.3665257692337036},{"id":"https://openalex.org/keywords/social-psychology","display_name":"Social psychology","score":0.35687971115112305},{"id":"https://openalex.org/keywords/econometrics","display_name":"Econometrics","score":0.33124250173568726},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.27949267625808716},{"id":"https://openalex.org/keywords/geography","display_name":"Geography","score":0.10138949751853943}],"concepts":[{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.7578318119049072},{"id":"https://openalex.org/C2779343474","wikidata":"https://www.wikidata.org/wiki/Q3109175","display_name":"Context (archaeology)","level":2,"score":0.6409345865249634},{"id":"https://openalex.org/C2777212361","wikidata":"https://www.wikidata.org/wiki/Q5127848","display_name":"Class (philosophy)","level":2,"score":0.6213930249214172},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.48993271589279175},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.4399090111255646},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.41161638498306274},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.3665257692337036},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.35687971115112305},{"id":"https://openalex.org/C149782125","wikidata":"https://www.wikidata.org/wiki/Q160039","display_name":"Econometrics","level":1,"score":0.33124250173568726},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.27949267625808716},{"id":"https://openalex.org/C205649164","wikidata":"https://www.wikidata.org/wiki/Q1071","display_name":"Geography","level":0,"score":0.10138949751853943},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0},{"id":"https://openalex.org/C166957645","wikidata":"https://www.wikidata.org/wiki/Q23498","display_name":"Archaeology","level":1,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.1458","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1458","pdf_url":"https://aclanthology.org/2025.acl-long.1458.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.1458","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1458","pdf_url":"https://aclanthology.org/2025.acl-long.1458.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":[{"score":0.550000011920929,"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities"}],"awards":[],"funders":[{"id":"https://openalex.org/F4320309480","display_name":"Nvidia","ror":"https://ror.org/03jdj4y14"},{"id":"https://openalex.org/F4320309949","display_name":"Canadian Institute for Advanced Research","ror":"https://ror.org/01sdtdd95"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412889746.pdf","grobid_xml":"https://content.openalex.org/works/W4412889746.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2748952813","https://openalex.org/W4391375266","https://openalex.org/W3162204513","https://openalex.org/W2371138613","https://openalex.org/W2048963458","https://openalex.org/W43109613","https://openalex.org/W2359952343","https://openalex.org/W2239445980","https://openalex.org/W2080152487","https://openalex.org/W3083152911"],"abstract_inverted_index":{"The":[0],"widespread":[1],"success":[2],"of":[3,39],"large":[4],"language":[5],"models":[6,58,87],"(LLMs)":[7],"on":[8,106,178,197],"NLP":[9],"benchmarks":[10],"has":[11],"been":[12],"accompanied":[13],"by":[14,151],"concerns":[15],"that":[16,23,70,80,118],"LLMs":[17,186],"function":[18],"primarily":[19],"as":[20],"stochastic":[21,180,203],"parrots":[22],"reproduce":[24],"texts":[25],"similar":[26],"to":[27,100],"what":[28,35],"they":[29],"saw":[30],"during":[31],"pre-training,":[32],"often":[33],"erroneously.But":[34],"is":[36,121,149],"the":[37,96,124,160,169,179],"nature":[38],"their":[40,64],"errors,":[41],"and":[42,109],"do":[43],"these":[44,71],"errors":[45,72],"exhibit":[46,188],"any":[47],"regularities?In":[48],"this":[49,119],"work,":[50],"we":[51,68,81,201],"examine":[52],"irrelevant":[53],"context":[54,99],"hallucinations,":[55],"in":[56,85,123,134,143,193],"which":[57,86],"integrate":[59],"misleading":[60],"contextual":[61,163,198],"cues":[62,91,164,199],"into":[63,140],"predictions.Through":[65],"behavioral":[66],"analysis,":[67],"show":[69],"result":[73],"from":[74,95],"a":[75,174],"structured":[76],"yet":[77],"flawed":[78],"mechanism":[79],"term":[82,202],"class-based":[83],"(mis)generalization,":[84],"combine":[88],"abstract":[89,129],"class":[90,130],"with":[92],"features":[93],"extracted":[94],"query":[97],"or":[98],"derive":[101],"answers.Furthermore,":[102],"mechanistic":[103],"interpretability":[104],"experiments":[105],"Llama-3,":[107],"Mistral,":[108],"Pythia":[110],"across":[111],"39":[112],"factual":[113],"recall":[114],"relation":[115],"types":[116],"reveal":[117],"behavior":[120],"reflected":[122],"model's":[125],"internal":[126],"computations:":[127],"(i)":[128],"representations":[131],"are":[132],"constructed":[133],"lower":[135],"layers":[136],"before":[137],"being":[138],"refined":[139],"specific":[141],"answers":[142],"higher":[144],"layers,":[145],"(ii)":[146],"feature":[147],"selection":[148],"governed":[150],"two":[152],"competing":[153],"circuits":[154],"-one":[155],"prioritizing":[156],"direct":[157],"query-based":[158],"reasoning,":[159],"other":[161],"incorporating":[162],"-whose":[165],"relative":[166],"influences":[167],"determine":[168],"final":[170],"output.Our":[171],"findings":[172],"provide":[173],"more":[175],"nuanced":[176],"perspective":[177],"parrot":[181],"argument:":[182],"through":[183],"form-based":[184],"training,":[185],"can":[187],"generalization":[189],"leveraging":[190],"abstractions,":[191],"albeit":[192],"unreliable":[194],"ways":[195],"based":[196],"-what":[200],"chameleons.":[204]},"counts_by_year":[{"year":2026,"cited_by_count":1}],"updated_date":"2026-06-11T09:08:48.828518","created_date":"2025-10-10T00:00:00"}