{"id":"https://openalex.org/W7161683761","doi":"https://doi.org/10.48550/arxiv.2605.18083","title":"A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$\u0394$ Integration into Upcycled MoE","display_name":"A Data-Efficient Path to Multilingual LLMs: Language Expansion via Post-training PARAM$\u0394$ Integration into Upcycled MoE","publication_year":2026,"publication_date":"2026-05-18","ids":{"openalex":"https://openalex.org/W7161683761","doi":"https://doi.org/10.48550/arxiv.2605.18083"},"language":null,"primary_location":{"id":"doi:10.48550/arxiv.2605.18083","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.18083","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"type":"preprint","indexed_in":["datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://doi.org/10.48550/arxiv.2605.18083","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5136463862","display_name":"Hao Zhou","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Zhou, Hao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136492460","display_name":"Tianhao Li","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Li, Tianhao","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136455049","display_name":"Zhijun Wang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wang, Zhijun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060798694","display_name":"Shuaijie She","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"She, Shuaijie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136482781","display_name":"Linjuan Wu","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wu, Linjuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136498559","display_name":"Hao-Ran Wei","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Wei, Hao-Ran","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5133581226","display_name":"Baosong Yang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Yang, Baosong","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5136479812","display_name":"Jiajun Chen","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Chen, Jiajun","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5136457822","display_name":"Shujian Huang","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Huang, Shujian","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":0,"citation_normalized_percentile":null,"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":null,"last_page":null},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.3937999904155731,"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":0.3937999904155731,"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":0.08590000122785568,"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/T13702","display_name":"Machine Learning in Healthcare","score":0.07109999656677246,"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/path","display_name":"Path (computing)","score":0.5595999956130981},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.47279998660087585},{"id":"https://openalex.org/keywords/base","display_name":"Base (topology)","score":0.4546000063419342},{"id":"https://openalex.org/keywords/language-understanding","display_name":"Language understanding","score":0.3483999967575073},{"id":"https://openalex.org/keywords/constructed-language","display_name":"Constructed language","score":0.30390000343322754},{"id":"https://openalex.org/keywords/cover","display_name":"Cover (algebra)","score":0.29589998722076416}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.767799973487854},{"id":"https://openalex.org/C2777735758","wikidata":"https://www.wikidata.org/wiki/Q817765","display_name":"Path (computing)","level":2,"score":0.5595999956130981},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.47279998660087585},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.46369999647140503},{"id":"https://openalex.org/C42058472","wikidata":"https://www.wikidata.org/wiki/Q810214","display_name":"Base (topology)","level":2,"score":0.4546000063419342},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.4251999855041504},{"id":"https://openalex.org/C199360897","wikidata":"https://www.wikidata.org/wiki/Q9143","display_name":"Programming language","level":1,"score":0.35830000042915344},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.3483999967575073},{"id":"https://openalex.org/C80444323","wikidata":"https://www.wikidata.org/wiki/Q2878974","display_name":"Theoretical computer science","level":1,"score":0.3361999988555908},{"id":"https://openalex.org/C94922259","wikidata":"https://www.wikidata.org/wiki/Q33215","display_name":"Constructed language","level":2,"score":0.30390000343322754},{"id":"https://openalex.org/C2780428219","wikidata":"https://www.wikidata.org/wiki/Q16952335","display_name":"Cover (algebra)","level":2,"score":0.29589998722076416},{"id":"https://openalex.org/C3826847","wikidata":"https://www.wikidata.org/wiki/Q188768","display_name":"FLOPS","level":2,"score":0.2824000120162964},{"id":"https://openalex.org/C165064840","wikidata":"https://www.wikidata.org/wiki/Q1321061","display_name":"Matching (statistics)","level":2,"score":0.2676999866962433},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.260699987411499}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.48550/arxiv.2605.18083","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.18083","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.48550/arxiv.2605.18083","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2605.18083","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I205783295","host_organization_name":"Cornell University","host_organization_lineage":["https://openalex.org/I205783295"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"},"sustainable_development_goals":[{"display_name":"Quality Education","id":"https://metadata.un.org/sdg/4","score":0.6781800389289856}],"awards":[],"funders":[],"has_content":{"pdf":false,"grobid_xml":false},"content_urls":null,"referenced_works_count":0,"referenced_works":[],"related_works":[],"abstract_inverted_index":{"Expanding":[0],"Large":[1],"Language":[2],"Models~(LLMs)":[3],"to":[4,24,47,77,91],"new":[5,53],"languages":[6,120],"is":[7,82,131],"a":[8,29,41,67,71,87],"costly":[9],"endeavor,":[10],"demanding":[11],"extensive":[12],"Continued":[13],"Pre-Training~(CPT)":[14],"and":[15,56,137],"data-intensive":[16],"alignment.":[17],"While":[18],"recent":[19],"data-free":[20],"merging":[21],"techniques":[22],"attempt":[23],"bypass":[25],"alignment":[26,99],"by":[27,40,85],"fusing":[28],"multilingual":[30],"CPT-enhanced":[31,93],"model":[32,69],"with":[33,108],"its":[34],"instruct":[35],"counterpart,":[36],"they":[37],"are":[38],"plagued":[39],"critical":[42],"trade-off:":[43],"mitigating":[44],"parameter":[45,89],"conflicts":[46],"preserve":[48],"original":[49,124],"abilities":[50],"inevitably":[51],"dilutes":[52],"language":[54],"acquisition,":[55],"vice-versa.":[57],"To":[58],"resolve":[59],"this":[60],"conflict,":[61],"we":[62],"introduce":[63],"\\method,":[64],"which":[65],"upcycles":[66],"dense":[68],"into":[70],"Mixture-of-Experts~(MoE)":[72],"architecture,":[73],"allocating":[74],"different":[75,78,135],"experts":[76],"languages.":[79],"Alignment":[80],"ability":[81],"then":[83],"transferred":[84],"grafting":[86],"MoE-expanded":[88],"delta~($\u0394_{\\text{post}}$)":[90],"the":[92,97],"base":[94],"model,":[95],"bypassing":[96],"complex":[98],"phase.":[100],"Experiments":[101],"demonstrate":[102],"\\method's":[103],"superiority":[104],"even":[105],"against":[106],"baselines":[107],"similar":[109],"FLOPs":[110],"or":[111],"number":[112],"of":[113],"parameters;":[114],"it":[115],"improves":[116],"performance":[117],"on":[118],"expanded":[119],"while":[121],"effectively":[122],"preserving":[123],"capabilities.":[125],"We":[126],"further":[127],"show":[128],"our":[129],"approach":[130],"highly":[132],"applicable":[133],"across":[134],"models":[136],"Post-training":[138],"deltas.":[139]},"counts_by_year":[],"updated_date":"2026-07-01T06:00:48.157686","created_date":"2026-05-20T00:00:00"}