{"id":"https://openalex.org/W7224445784","doi":"https://doi.org/10.18653/2026.findings-emnlp.1944","title":"Sparse Long Memory and Transition Bottlenecks Shape Positional Recall in Mamba Language Models","display_name":"Sparse Long Memory and Transition Bottlenecks Shape Positional Recall in Mamba Language Models","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7224445784","doi":"https://doi.org/10.18653/2026.findings-emnlp.1944"},"language":"en","primary_location":{"id":"doi:10.18653/2026.findings-emnlp.1944","doi":"https://doi.org/10.18653/2026.findings-emnlp.1944","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.findings-emnlp.1944","pdf_url":"https://aclanthology.org/2026.findings-emnlp.1944.pdf","source":{"id":"https://openalex.org/S7407096085","display_name":"Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2026","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","is_traditional":true,"indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://aclanthology.org/2026.findings-emnlp.1944.pdf","any_repository_has_fulltext":false},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5137652185","display_name":"Muhammad Cendekia Airlangga","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Muhammad Cendekia Airlangga","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5025882033","display_name":"Hilal AlQuabeh","orcid":"https://orcid.org/0000-0002-7438-3185","observed_orcids":["https://orcid.org/0000-0002-7438-3185"]},"institutions":[{"id":"https://openalex.org/I4210110652","display_name":"RIKEN","ror":"https://ror.org/01sjwvz98","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Hilal AlQuabeh","raw_affiliation_strings":["RIKEN ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"RIKEN ,","institution_ids":["https://openalex.org/I4210110652"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5034207626","display_name":"Munachiso Samuel Nwadike","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4210110652","display_name":"RIKEN","ror":"https://ror.org/01sjwvz98","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Munachiso Samuel Nwadike","raw_affiliation_strings":["RIKEN ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"RIKEN ,","institution_ids":["https://openalex.org/I4210110652"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5101815181","display_name":"Kentaro Inui","orcid":"https://orcid.org/0000-0001-6510-604X","observed_orcids":["https://orcid.org/0000-0001-6510-604X"]},"institutions":[{"id":"https://openalex.org/I201537933","display_name":"Tohoku University","ror":"https://ror.org/01dq60k83","country_code":"JP","type":"education","lineage":["https://openalex.org/I201537933"]},{"id":"https://openalex.org/I4210110652","display_name":"RIKEN","ror":"https://ror.org/01sjwvz98","country_code":"JP","type":"facility","lineage":["https://openalex.org/I4210110652"]}],"countries":["JP"],"is_corresponding":false,"raw_author_name":"Kentaro Inui","raw_affiliation_strings":["RIKEN ,","Tohoku University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"RIKEN ,","institution_ids":["https://openalex.org/I4210110652"]},{"raw_affiliation_string":"Tohoku University","institution_ids":["https://openalex.org/I201537933"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":0.0,"has_fulltext":true,"cited_by_count":0,"citation_normalized_percentile":{"value":0.76242855,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"41448","last_page":"41463"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.7401999831199646,"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":0.7401999831199646,"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":0.08240000158548355,"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":0.027799999341368675,"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/long-term-memory","display_name":"long-term memory","score":0.4269999861717224},{"id":"https://openalex.org/keywords/recency","display_name":"recency","score":0.34299999475479126},{"id":"https://openalex.org/keywords/primacy","display_name":"primacy","score":0.7390000224113464},{"id":"https://openalex.org/keywords/long-context-understanding","display_name":"long-context understanding","score":0.4659999907016754},{"id":"https://openalex.org/keywords/summarization","display_name":"summarization","score":0.7099999785423279},{"id":"https://openalex.org/keywords/question-answering","display_name":"question answering","score":0.48100000619888306}],"concepts":[{"id":"https://openalex.org/C100660578","wikidata":"https://www.wikidata.org/wiki/Q18733","display_name":"Recall","level":2,"score":0.5683000087738037},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.5586000084877014},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4862000048160553},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.413100004196167},{"id":"https://openalex.org/C194232998","wikidata":"https://www.wikidata.org/wiki/Q1606712","display_name":"Transition (genetics)","level":3,"score":0.3824999928474426},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.3637999892234802},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.3043000102043152},{"id":"https://openalex.org/C12186640","wikidata":"https://www.wikidata.org/wiki/Q6815743","display_name":"Memory model","level":3,"score":0.2883000075817108},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.26930001378059387},{"id":"https://openalex.org/C81669768","wikidata":"https://www.wikidata.org/wiki/Q2359161","display_name":"Precision and recall","level":2,"score":0.2565000057220459},{"id":"https://openalex.org/C153180895","wikidata":"https://www.wikidata.org/wiki/Q7148389","display_name":"Pattern recognition (psychology)","level":2,"score":0.25200000405311584}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/2026.findings-emnlp.1944","doi":"https://doi.org/10.18653/2026.findings-emnlp.1944","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.findings-emnlp.1944","pdf_url":"https://aclanthology.org/2026.findings-emnlp.1944.pdf","source":{"id":"https://openalex.org/S7407096085","display_name":"Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2026","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/2026.findings-emnlp.1944","doi":"https://doi.org/10.18653/2026.findings-emnlp.1944","is_oa":true,"landing_page_url":"https://doi.org/10.18653/2026.findings-emnlp.1944","pdf_url":"https://aclanthology.org/2026.findings-emnlp.1944.pdf","source":{"id":"https://openalex.org/S7407096085","display_name":"Findings of the Association for Computational Linguistics: EMNLP (Findings of EMNLP)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Findings of the Association for Computational Linguistics: EMNLP 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7224445784.pdf","grobid_xml":"https://content.openalex.org/works/W7224445784.grobid-xml"},"referenced_works_count":5,"referenced_works":["https://openalex.org/W3209374680","https://openalex.org/W4297412003","https://openalex.org/W4388514667","https://openalex.org/W4403346143","https://openalex.org/W4417284069"],"related_works":["https://openalex.org/W7158755242","https://openalex.org/W7138838935","https://openalex.org/W7158897782","https://openalex.org/W7172410052","https://openalex.org/W4412888538","https://openalex.org/W7224336409","https://openalex.org/W7164029421","https://openalex.org/W4415445242","https://openalex.org/W7137837368","https://openalex.org/W4415333514"],"abstract_inverted_index":{"Mamba":[0,35],"language":[1],"models":[2],"process":[3],"sequences":[4],"recurrently,":[5],"but":[6,111],"recurrent":[7],"processing":[8],"does":[9],"not":[10,114],"imply":[11],"that":[12,25,34,44,53,75],"all":[13],"tokens":[14],"are":[15,88,103,112],"retained":[16],"equally.We":[17],"study":[18],"this":[19,60,124],"gap":[20],"using":[21],"structured":[22],"recall":[23,66,140],"tasks":[24],"expose":[26],"primacy,":[27],"recency,":[28,46],"and":[29,47,67,72,150],"lost-in-themiddle":[30],"behavior.Our":[31],"results":[32,94],"suggest":[33],"uses":[36],"two":[37],"functional":[38],"memory":[39],"regimes:":[40],"a":[41,48,96,134,151],"short-lived":[42],"state":[43],"supports":[45],"sparse":[49,118],"long-lived":[50,119],"channel":[51],"subspace":[52,62],"preserves":[54],"selected":[55],"early":[56,65],"inputs.Targeted":[57],"interventions":[58],"on":[59],"long-memory":[61],"selectively":[63],"degrade":[64],"transfer":[68],"to":[69],"long-context":[70],"QA":[71],"summarization,":[73],"indicating":[74],"the":[76,117,128,143],"same":[77],"channels":[78],"support":[79,95],"real":[80],"longcontext":[81],"use.We":[82],"then":[83],"ask":[84],"why":[85],"middle":[86,101],"items":[87,102],"forgotten":[89],"despite":[90],"being":[91],"processed":[92],"sequentially.Our":[93],"short-to-long":[97],"transition":[98,153],"bottleneck":[99,125],"account:":[100],"no":[104],"longer":[105],"recent":[106],"enough":[107],"for":[108],"short":[109,148],"memory,":[110,147,149],"often":[113],"promoted":[115],"into":[116],"subspace.Repeated":[120],"input":[121],"structure":[122],"strengthens":[123],"by":[126],"changing":[127],"input-dependent":[129],"discretization":[130],"dynamics.These":[131],"findings":[132],"provide":[133],"mechanistic":[135],"account":[136],"of":[137],"Mamba's":[138],"U-shaped":[139],"curve":[141],"as":[142],"interaction":[144],"between":[145,154],"long":[146],"fragile":[152],"them.":[155]},"counts_by_year":[],"updated_date":"2026-11-08T12:12:28.789561","created_date":"2026-10-30T00:00:00"}