{"id":"https://openalex.org/W4416035582","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1448","title":"Do All Autoregressive Transformers Remember Facts the Same Way? A Cross-Architecture Analysis of Recall Mechanisms","display_name":"Do All Autoregressive Transformers Remember Facts the Same Way? A Cross-Architecture Analysis of Recall Mechanisms","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4416035582","doi":"https://doi.org/10.18653/v1/2025.emnlp-main.1448"},"language":null,"primary_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1448","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1448","pdf_url":"https://aclanthology.org/2025.emnlp-main.1448.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"type":"article","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.emnlp-main.1448.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5120308865","display_name":"Minyeong Choe","orcid":null},"institutions":[],"countries":[],"is_corresponding":true,"raw_author_name":"Minyeong Choe","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060286024","display_name":"Haehyun Cho","orcid":"https://orcid.org/0000-0002-5344-5252"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Haehyun Cho","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5010473957","display_name":"Changho Seo","orcid":"https://orcid.org/0000-0002-0779-3539"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Changho Seo","raw_affiliation_strings":[],"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5035508458","display_name":"Hyunil Kim","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Hyunil Kim","raw_affiliation_strings":[],"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5120308865"],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.38583206,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"28482","last_page":"28501"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10525","display_name":"Human-Automation Interaction and Safety","score":0.04500000178813934,"subfield":{"id":"https://openalex.org/subfields/3207","display_name":"Social Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},"topics":[{"id":"https://openalex.org/T10525","display_name":"Human-Automation Interaction and Safety","score":0.04500000178813934,"subfield":{"id":"https://openalex.org/subfields/3207","display_name":"Social Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T11148","display_name":"Language, Metaphor, and Cognition","score":0.023800000548362732,"subfield":{"id":"https://openalex.org/subfields/3205","display_name":"Experimental and Cognitive Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10970","display_name":"Team Dynamics and Performance","score":0.021299999207258224,"subfield":{"id":"https://openalex.org/subfields/3207","display_name":"Social Psychology"},"field":{"id":"https://openalex.org/fields/32","display_name":"Psychology"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/recall","display_name":"Recall","score":0.677299976348877},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.46160000562667847},{"id":"https://openalex.org/keywords/autoregressive-model","display_name":"Autoregressive model","score":0.3889999985694885},{"id":"https://openalex.org/keywords/precision-and-recall","display_name":"Precision and recall","score":0.25699999928474426}],"concepts":[{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.677299976348877},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.47600001096725464},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.46160000562667847},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.3937999904155731},{"id":"https://openalex.org/C159877910","wikidata":"https://www.wikidata.org/wiki/Q2202883","display_name":"Autoregressive model","level":2,"score":0.3889999985694885},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.3352999985218048},{"id":"https://openalex.org/C180747234","wikidata":"https://www.wikidata.org/wiki/Q23373","display_name":"Cognitive psychology","level":1,"score":0.30640000104904175},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.2754000127315521},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.2752000093460083},{"id":"https://openalex.org/C81669768","wikidata":"https://www.wikidata.org/wiki/Q2359161","display_name":"Precision and recall","level":2,"score":0.25699999928474426}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.emnlp-main.1448","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1448","pdf_url":"https://aclanthology.org/2025.emnlp-main.1448.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.emnlp-main.1448","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.emnlp-main.1448","pdf_url":"https://aclanthology.org/2025.emnlp-main.1448.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 2025 Conference on Empirical Methods in Natural Language Processing","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G6072120315","display_name":null,"funder_award_id":"funded","funder_id":"https://openalex.org/F4320335489","funder_display_name":"Institute for Information and Communications Technology Promotion"},{"id":"https://openalex.org/G7690800436","display_name":null,"funder_award_id":"RS-2024-00398353","funder_id":"https://openalex.org/F4320328359","funder_display_name":"Ministry of Science and ICT, South Korea"}],"funders":[{"id":"https://openalex.org/F4320328359","display_name":"Ministry of Science and ICT, South Korea","ror":"https://ror.org/01wpjm123"},{"id":"https://openalex.org/F4320335489","display_name":"Institute for Information and Communications Technology Promotion","ror":"https://ror.org/01g0hqq23"}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4416035582.pdf","grobid_xml":"https://content.openalex.org/works/W4416035582.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Understanding":[0],"how":[1,69],"Transformer-based":[2],"language":[3],"models":[4,80],"store":[5],"and":[6,15,65,68,74],"retrieve":[7],"factual":[8,36,57,70,95,118],"associations":[9],"is":[10,72],"critical":[11],"for":[12],"improving":[13],"interpretability":[14],"enabling":[16],"targeted":[17],"model":[18],"editing.Prior":[19],"work,":[20],"primarily":[21],"on":[22],"GPT-style":[23],"models,":[24],"has":[25],"identified":[26],"MLP":[27,98],"modules":[28,87],"in":[29,88],"early":[30],"layers":[31,91],"as":[32],"key":[33],"contributors":[34],"to":[35,94,113],"recall.However,":[37],"it":[38],"remains":[39],"unclear":[40],"whether":[41],"these":[42],"findings":[43,100],"generalize":[44],"across":[45,59],"different":[46,115],"autoregressive":[47,106],"architectures.To":[48],"address":[49],"this,":[50],"we":[51,76],"conduct":[52],"a":[53],"comprehensive":[54],"evaluation":[55],"of":[56,117],"recall":[58,96],"several":[60],"models-including":[61],"GPT,":[62],"LLaMA,":[63],"Qwen,":[64],"DeepSeek-analyzing":[66],"where":[67],"information":[71],"encoded":[73],"accessed.Consequently,":[75],"find":[77],"that":[78,102],"Qwen-based":[79],"behave":[81],"differently":[82],"from":[83],"previous":[84],"patterns:":[85],"attention":[86],"the":[89,105],"earliest":[90],"contribute":[92],"more":[93],"than":[97],"modules.Our":[99],"suggest":[101],"even":[103],"within":[104],"Transformer":[107],"family,":[108],"architectural":[109],"variations":[110],"can":[111],"lead":[112],"fundamentally":[114],"mechanisms":[116],"recall.":[119]},"counts_by_year":[],"updated_date":"2026-03-25T14:56:36.534964","created_date":"2025-11-08T00:00:00"}