{"id":"https://openalex.org/W7140144739","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.286","title":"Improving the OOD Performance of Closed-Source LLMs on NLI Through Strategic Data Selection","display_name":"Improving the OOD Performance of Closed-Source LLMs on NLI Through Strategic Data Selection","publication_year":2026,"publication_date":"2026-01-01","ids":{"openalex":"https://openalex.org/W7140144739","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.286"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2026.findings-eacl.286","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.286","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.286","pdf_url":"https://aclanthology.org/2026.findings-eacl.286.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"hybrid","oa_url":"https://aclanthology.org/2026.findings-eacl.286.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5078650400","display_name":"Joe Stacey","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I47508984","display_name":"Imperial College London","ror":"https://ror.org/041kmwe10","country_code":"GB","type":"education","lineage":["https://openalex.org/I47508984"]},{"id":"https://openalex.org/I91136226","display_name":"University of Sheffield","ror":"https://ror.org/05krs5044","country_code":"GB","type":"education","lineage":["https://openalex.org/I91136226"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Joe Stacey","raw_affiliation_strings":["Imperial College London ,","University of Sheffield ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Imperial College London ,","institution_ids":["https://openalex.org/I47508984"]},{"raw_affiliation_string":"University of Sheffield ,","institution_ids":["https://openalex.org/I91136226"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5001337463","display_name":"Lisa Alazraki","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I47508984","display_name":"Imperial College London","ror":"https://ror.org/041kmwe10","country_code":"GB","type":"education","lineage":["https://openalex.org/I47508984"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Lisa Alazraki","raw_affiliation_strings":["Imperial College London ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Imperial College London ,","institution_ids":["https://openalex.org/I47508984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5119756980","display_name":"Aran Ubhi","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I47508984","display_name":"Imperial College London","ror":"https://ror.org/041kmwe10","country_code":"GB","type":"education","lineage":["https://openalex.org/I47508984"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Aran Ubhi","raw_affiliation_strings":["Imperial College London ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Imperial College London ,","institution_ids":["https://openalex.org/I47508984"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5057377890","display_name":"Beyza Ermi\u015f","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I4401726847","display_name":"Cohere (Canada)","ror":"https://ror.org/050yr0479","country_code":"US","type":"company","lineage":["https://openalex.org/I4401726847"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Beyza Ermis","raw_affiliation_strings":["Cohere Labs ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Cohere Labs ,","institution_ids":["https://openalex.org/I4401726847"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5139844625","display_name":"Aaron Mueller","orcid":"https://orcid.org/0009-0005-1148-5001","observed_orcids":["https://orcid.org/0009-0005-1148-5001"]},"institutions":[{"id":"https://openalex.org/I111088046","display_name":"Boston University","ror":"https://ror.org/05qwgg493","country_code":"US","type":"education","lineage":["https://openalex.org/I111088046"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Aaron Mueller","raw_affiliation_strings":["Boston University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Boston University","institution_ids":["https://openalex.org/I111088046"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5062209641","display_name":"Marek Rei","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I47508984","display_name":"Imperial College London","ror":"https://ror.org/041kmwe10","country_code":"GB","type":"education","lineage":["https://openalex.org/I47508984"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Marek Rei","raw_affiliation_strings":["Imperial College London ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Imperial College London ,","institution_ids":["https://openalex.org/I47508984"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"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":0.15648904,"is_in_top_1_percent":false,"is_in_top_10_percent":false},"cited_by_percentile_year":null,"biblio":{"volume":null,"issue":null,"first_page":"5378","last_page":"5404"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.8974999785423279,"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":0.8974999785423279,"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":0.042100001126527786,"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/T11689","display_name":"Adversarial Robustness in Machine Learning","score":0.030500000342726707,"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/natural-language-inference","display_name":"natural language inference","score":0.8820000290870667},{"id":"https://openalex.org/keywords/out-of-distribution-robustness","display_name":"out-of-distribution robustness","score":0.9580000042915344},{"id":"https://openalex.org/keywords/fine-tuning","display_name":"fine-tuning","score":0.7179999947547913},{"id":"https://openalex.org/keywords/training-sample-selection","display_name":"training data selection","score":0.8379999995231628},{"id":"https://openalex.org/keywords/synthetic-training-data","display_name":"synthetic training data","score":0.7239999771118164},{"id":"https://openalex.org/keywords/llm-generated-data","display_name":"LLM-generated data","score":0.8009999990463257},{"id":"https://openalex.org/keywords/distribution-shift","display_name":"distribution shift","score":0.8740000128746033},{"id":"https://openalex.org/keywords/autoregressive-language-models","display_name":"autoregressive language models","score":0.7360000014305115},{"id":"https://openalex.org/keywords/encoder-models","display_name":"encoder models","score":0.671999990940094}],"concepts":[{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5358999967575073},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.4577000141143799},{"id":"https://openalex.org/C144133560","wikidata":"https://www.wikidata.org/wiki/Q4830453","display_name":"Business","level":0,"score":0.43849998712539673},{"id":"https://openalex.org/C48243021","wikidata":"https://www.wikidata.org/wiki/Q932522","display_name":"Strategic planning","level":2,"score":0.3720000088214874},{"id":"https://openalex.org/C112930515","wikidata":"https://www.wikidata.org/wiki/Q4389547","display_name":"Risk analysis (engineering)","level":1,"score":0.3594000041484833},{"id":"https://openalex.org/C195094911","wikidata":"https://www.wikidata.org/wiki/Q14167904","display_name":"Process management","level":1,"score":0.34860000014305115},{"id":"https://openalex.org/C162118730","wikidata":"https://www.wikidata.org/wiki/Q1128453","display_name":"Actuarial science","level":1,"score":0.28859999775886536},{"id":"https://openalex.org/C26517878","wikidata":"https://www.wikidata.org/wiki/Q228039","display_name":"Key (lock)","level":2,"score":0.2809999883174896},{"id":"https://openalex.org/C2778137410","wikidata":"https://www.wikidata.org/wiki/Q2732820","display_name":"Government (linguistics)","level":2,"score":0.27250000834465027},{"id":"https://openalex.org/C21547014","wikidata":"https://www.wikidata.org/wiki/Q1423657","display_name":"Operations management","level":1,"score":0.2655999958515167},{"id":"https://openalex.org/C162324750","wikidata":"https://www.wikidata.org/wiki/Q8134","display_name":"Economics","level":0,"score":0.2646999955177307},{"id":"https://openalex.org/C56739046","wikidata":"https://www.wikidata.org/wiki/Q192060","display_name":"Knowledge management","level":1,"score":0.2628999948501587},{"id":"https://openalex.org/C42475967","wikidata":"https://www.wikidata.org/wiki/Q194292","display_name":"Operations research","level":1,"score":0.2605000138282776},{"id":"https://openalex.org/C98045186","wikidata":"https://www.wikidata.org/wiki/Q205663","display_name":"Process (computing)","level":2,"score":0.2574000060558319}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2026.findings-eacl.286","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.286","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.286","pdf_url":"https://aclanthology.org/2026.findings-eacl.286.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2505.20209","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2505.20209","pdf_url":"https://arxiv.org/pdf/2505.20209","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2505.20209","doi":"https://doi.org/10.48550/arxiv.2505.20209","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2505.20209","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":"2331-8422","issn":["2331-8422"],"is_oa":true,"is_in_doaj":false,"is_core":false,"listed_in":[],"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":null,"license_id":null,"version":null,"is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"Preprint"}],"best_oa_location":{"id":"doi:10.18653/v1/2026.findings-eacl.286","doi":"https://doi.org/10.18653/v1/2026.findings-eacl.286","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2026.findings-eacl.286","pdf_url":"https://aclanthology.org/2026.findings-eacl.286.pdf","source":{"id":"https://openalex.org/S7407087051","display_name":"Findings of the Association for Computational Linguistics: EACL","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":[],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"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: EACL 2026","raw_type":"proceedings-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[{"score":0.41896572709083557,"id":"https://metadata.un.org/sdg/9","display_name":"Industry, innovation and infrastructure"}],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7140144739.pdf","grobid_xml":"https://content.openalex.org/works/W7140144739.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4379958578","https://openalex.org/W4402671976","https://openalex.org/W4402671047","https://openalex.org/W7117106895","https://openalex.org/W4411120517","https://openalex.org/W7036295773","https://openalex.org/W7127417789","https://openalex.org/W7163209585","https://openalex.org/W4386566613","https://openalex.org/W4394867291"],"abstract_inverted_index":{"We":[0],"investigate":[1],"the":[2,11,20,34,60,83],"robustness":[3,43,56],"of":[4,13,37],"fine-tuned":[5],"Large":[6],"Language":[7,15],"Models":[8],"(LLMs)":[9],"for":[10,74,175],"task":[12],"Natural":[14],"Inference":[16],"(NLI),":[17],"finding":[18],"that":[19,46,68,124,154],"in-distribution":[21],"gains":[22],"from":[23],"fine-tuning":[24,51,61,85],"correspond":[25],"to":[26,54,116,135,162,166],"a":[27,172],"large":[28],"drop":[29],"in":[30],"out-of-distribution":[31],"(OOD)":[32],"performance.Despite":[33],"widespread":[35],"use":[36],"closed-source":[38,75],"LLMs,":[39],"there":[40],"are":[41,69,127,158],"no":[42],"mitigation":[44],"methods":[45,53,67],"work":[47],"under":[48],"their":[49],"API":[50],"constraints.Existing":[52],"improve":[55,143],"typically":[57],"require":[58],"changing":[59],"process":[62],"or":[63,71,91],"large-scale":[64],"data":[65,114,140],"augmentation,":[66],"infeasible":[70],"cost":[72],"prohibitive":[73],"models.To":[76],"address":[77],"this,":[78],"we":[79,141,152],"propose":[80],"strategically":[81],"selecting":[82],"NLI":[84,108],"data,":[86],"prioritising":[87],"more":[88,99,137,160],"complex":[89,100,138],"examples":[90,95,102,126],"replacing":[92],"existing":[93],"training":[94,101,111],"with":[96,112],"LLM-generated":[97],"data.Prioritising":[98],"improves":[103],"performance":[104,144],"on":[105,119,145],"challenging":[106,149],"OOD":[107,121,150],"datasets,":[109],"while":[110],"synthetic":[113,125,139],"leads":[115],"substantial":[117],"improvements":[118],"easier":[120],"datasets.We":[122],"find":[123],"often":[128],"too":[129],"simple,":[130],"and":[131,148,169],"by":[132],"prompting":[133],"LLMs":[134,157],"create":[136],"can":[142],"both":[146],"easy":[147],"datasets.Finally,":[151],"show":[153],"recent":[155],"autoregressive":[156],"substantially":[159],"robust":[161],"distributional":[163],"shifts":[164],"compared":[165],"encoder":[167],"models,":[168],"should":[170],"be":[171],"preferred":[173],"baseline":[174],"future":[176],"research.1":[177]},"counts_by_year":[],"updated_date":"2026-10-27T15:51:10.671819","created_date":"2026-03-24T00:00:00"}