{"id":"https://openalex.org/W4412944494","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1354","title":"RomanLens: The Role Of Latent Romanization In Multilinguality In LLMs","display_name":"RomanLens: The Role Of Latent Romanization In Multilinguality In LLMs","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412944494","doi":"https://doi.org/10.18653/v1/2025.findings-acl.1354"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.1354","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1354","pdf_url":"https://aclanthology.org/2025.findings-acl.1354.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":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.1354.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5116248987","display_name":"Alan Saji","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Alan Saji","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5093800118","display_name":"Jaavid Aktar Husain","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jaavid Aktar Husain","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5104199197","display_name":"Thanmay Jayakumar","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Thanmay Jayakumar","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5077362985","display_name":"Raj Dabre","orcid":"https://orcid.org/0000-0003-0664-3421"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Raj Dabre","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5052909911","display_name":"Anoop Kunchukuttan","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Anoop Kunchukuttan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5069271622","display_name":"Ratish Puduppully","orcid":"https://orcid.org/0009-0001-0857-2842"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ratish Puduppully","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":false,"cited_by_count":2,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"26410","last_page":"26429"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9940999746322632,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9940999746322632,"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/T12881","display_name":"linguistics and terminology studies","score":0.9492999911308289,"subfield":{"id":"https://openalex.org/subfields/1203","display_name":"Language and Linguistics"},"field":{"id":"https://openalex.org/fields/12","display_name":"Arts and Humanities"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10759","display_name":"Translation Studies and Practices","score":0.9372000098228455,"subfield":{"id":"https://openalex.org/subfields/1203","display_name":"Language and Linguistics"},"field":{"id":"https://openalex.org/fields/12","display_name":"Arts and Humanities"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}}],"keywords":[{"id":"https://openalex.org/keywords/romanization","display_name":"Romanization","score":0.9039629697799683},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.430259108543396},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.33975768089294434},{"id":"https://openalex.org/keywords/history","display_name":"History","score":0.30598992109298706},{"id":"https://openalex.org/keywords/ancient-history","display_name":"Ancient history","score":0.18691593408584595}],"concepts":[{"id":"https://openalex.org/C106930687","wikidata":"https://www.wikidata.org/wiki/Q976327","display_name":"Romanization","level":2,"score":0.9039629697799683},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.430259108543396},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.33975768089294434},{"id":"https://openalex.org/C95457728","wikidata":"https://www.wikidata.org/wiki/Q309","display_name":"History","level":0,"score":0.30598992109298706},{"id":"https://openalex.org/C195244886","wikidata":"https://www.wikidata.org/wiki/Q41493","display_name":"Ancient history","level":1,"score":0.18691593408584595}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.1354","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1354","pdf_url":"https://aclanthology.org/2025.findings-acl.1354.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":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},{"id":"pmh:oai:pure.atira.dk:publications/463dd0de-336f-42cc-9dba-b7dc1420a77d","is_oa":false,"landing_page_url":"https://pure.itu.dk/portal/da/publications/463dd0de-336f-42cc-9dba-b7dc1420a77d","pdf_url":null,"source":{"id":"https://openalex.org/S4377196680","display_name":"IT University Of Copenhagen (IT University of Copenhagen)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I83467386","host_organization_name":"IT University of Copenhagen","host_organization_lineage":["https://openalex.org/I83467386"],"host_organization_lineage_names":[],"type":"repository"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Saji, A, Husain, J A, Jayakumar, T, Dabre, R, Kunchukuttan, A & Puduppully, R 2025, RomanLens: The Role Of Latent Romanization In Multilinguality In LLMs. in Findings of the Association for Computational Linguistics: ACL 2025. vol. Findings of the Association for Computational Linguistics: ACL 2025, Association for Computational Linguistics, Vienna, Austria, Findings of the Association for Computational Linguistics 2025, Annual Meeting of the Association for Computational Linguistics, Vienna, Austria, 27/07/2025. https://doi.org/10.18653/v1/2025.findings-acl.1354","raw_type":"info:eu-repo/semantics/conferenceObject"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.1354","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.1354","pdf_url":"https://aclanthology.org/2025.findings-acl.1354.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":"Findings of the Association for Computational Linguistics: ACL 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/10","display_name":"Reduced inequalities","score":0.49000000953674316}],"awards":[],"funders":[],"has_content":{"grobid_xml":false,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412944494.pdf"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4391375266","https://openalex.org/W2899084033","https://openalex.org/W2748952813","https://openalex.org/W2170685659","https://openalex.org/W2899224223","https://openalex.org/W2376933048","https://openalex.org/W4238918564","https://openalex.org/W2294494798","https://openalex.org/W3184806679","https://openalex.org/W1974228456"],"abstract_inverted_index":{"Large":[0],"Language":[1],"Models":[2],"(LLMs)":[3],"exhibit":[4],"strong":[5],"multilingual":[6,25,51,146],"performance":[7],"despite":[8],"being":[9],"predominantly":[10],"trained":[11],"on":[12,28],"English-centric":[13],"corpora.":[14],"This":[15],"raises":[16],"a":[17,47,78,103,142],"fundamental":[18],"question:":[19],"How":[20],"do":[21],"LLMs":[22,92,149],"achieve":[23],"such":[24],"capabilities?":[26],"Focusing":[27],"languages":[29],"written":[30],"in":[31,50,70,123,130,148,157],"non-Roman":[32,42,111],"scripts,":[33,101],"we":[34,57,80,89],"investigate":[35],"the":[36,119,131,152],"role":[37,154],"of":[38,41,145,155],"Romanization\u2014the":[39],"representation":[40,147],"scripts":[43],"using":[44],"Roman":[45],"characters\u2014as":[46],"potential":[48],"bridge":[49],"processing.":[52],"Using":[53],"mechanistic":[54],"interpretability":[55],"techniques,":[56],"analyze":[58],"next-token":[59],"generation":[60],"and":[61,99,150],"find":[62],"that":[63,91,117],"intermediate":[64],"layers":[65,133],"frequently":[66],"represent":[67],"target":[68,120],"words":[69],"Romanized":[71,100,124],"form":[72],"before":[73],"transitioning":[74],"to":[75,135,141],"native":[76,98,136],"script,":[77],"phenomenon":[79],"term":[81],"Latent":[82],"Romanization.":[83],"Further,":[84],"through":[85],"activation":[86],"patching":[87],"experiments,":[88],"demonstrate":[90],"encode":[93],"semantic":[94],"concepts":[95],"similarly":[96],"across":[97],"suggesting":[102],"shared":[104],"underlying":[105],"representation.":[106],"Additionally,":[107],"for":[108],"translation":[109],"into":[110],"script":[112],"languages,":[113],"our":[114],"findings":[115],"reveal":[116],"when":[118],"language":[121,159],"is":[122],"form,":[125],"its":[126],"representations":[127],"emerge":[128],"earlier":[129],"model\u2019s":[132],"compared":[134],"script.":[137],"These":[138],"insights":[139],"contribute":[140],"deeper":[143],"understanding":[144],"highlight":[151],"implicit":[153],"Romanization":[156],"facilitating":[158],"transfer.":[160]},"counts_by_year":[{"year":2026,"cited_by_count":2}],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}
