{"id":"https://openalex.org/W7128782817","doi":"https://doi.org/10.48550/arxiv.2602.12005","title":"LaCy: What Small Language Models Can and Should Learn is Not Just a Question of Loss","display_name":"LaCy: What Small Language Models Can and Should Learn is Not Just a Question of Loss","publication_year":2026,"publication_date":"2026-02-12","ids":{"openalex":"https://openalex.org/W7128782817","doi":"https://doi.org/10.48550/arxiv.2602.12005"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2602.12005","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2602.12005","pdf_url":"https://arxiv.org/pdf/2602.12005","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":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"volume":null,"type":"preprint","indexed_in":["arxiv","datacite"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2602.12005","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5027803852","display_name":"Szilvia Ujv\u00e1ry","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Ujv\u00e1ry, Szilvia","raw_affiliation_strings":["University of Cambridge ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Cambridge ,","institution_ids":["https://openalex.org/I241749"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5060903205","display_name":"Louis B\u00e9thune","orcid":"https://orcid.org/0000-0003-1498-8251","observed_orcids":["https://orcid.org/0000-0003-1498-8251"]},"institutions":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"B\u00e9thune, Louis","raw_affiliation_strings":["University of Cambridge ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Cambridge ,","institution_ids":["https://openalex.org/I241749"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5042340163","display_name":"Pierre Ablin","orcid":"https://orcid.org/0000-0003-4277-5202","observed_orcids":["https://orcid.org/0000-0003-4277-5202"]},"institutions":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Ablin, Pierre","raw_affiliation_strings":["University of Cambridge ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Cambridge ,","institution_ids":["https://openalex.org/I241749"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5125951156","display_name":"Jo\u00e3o Monteiro","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Monteiro, Jo\u00e3o","raw_affiliation_strings":["University of Cambridge ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Cambridge ,","institution_ids":["https://openalex.org/I241749"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5017263427","display_name":"Marco Cuturi","orcid":"https://orcid.org/0000-0002-1934-0588","observed_orcids":["https://orcid.org/0000-0002-1934-0588"]},"institutions":[{"id":"https://openalex.org/I241749","display_name":"University of Cambridge","ror":"https://ror.org/013meh722","country_code":"GB","type":"education","lineage":["https://openalex.org/I241749"]}],"countries":["GB"],"is_corresponding":false,"raw_author_name":"Cuturi, Marco","raw_affiliation_strings":["University of Cambridge ,"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Cambridge ,","institution_ids":["https://openalex.org/I241749"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5125922479","display_name":"Michael Kirchhof","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Kirchhof, Michael","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":null,"has_fulltext":true,"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":0.9150000214576721,"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":0.9150000214576721,"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":0.07419999688863754,"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/T12031","display_name":"Speech and dialogue systems","score":0.002099999925121665,"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/small-language-models","display_name":"small language models","score":0.996999979019165},{"id":"https://openalex.org/keywords/factuality","display_name":"factuality","score":0.4269999861717224},{"id":"https://openalex.org/keywords/pretraining","display_name":"pre-training","score":0.44999998807907104},{"id":"https://openalex.org/keywords/grammatical-information","display_name":"grammatical information","score":0.7459999918937683},{"id":"https://openalex.org/keywords/wikipedia","display_name":"Wikipedia","score":0.38499999046325684},{"id":"https://openalex.org/keywords/factual-hallucination","display_name":"factual hallucination","score":0.40400001406669617},{"id":"https://openalex.org/keywords/lactose-permease","display_name":"lactose permease","score":0.45399999618530273}],"concepts":[{"id":"https://openalex.org/C143273055","wikidata":"https://www.wikidata.org/wiki/Q2382794","display_name":"Delegate","level":2,"score":0.8819000124931335},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7968999743461609},{"id":"https://openalex.org/C48145219","wikidata":"https://www.wikidata.org/wiki/Q1335365","display_name":"Security token","level":2,"score":0.7059999704360962},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.5756999850273132},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.5680000185966492},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5601999759674072},{"id":"https://openalex.org/C186644900","wikidata":"https://www.wikidata.org/wiki/Q194152","display_name":"Parsing","level":2,"score":0.54339998960495},{"id":"https://openalex.org/C81917197","wikidata":"https://www.wikidata.org/wiki/Q628760","display_name":"Selection (genetic algorithm)","level":2,"score":0.5291000008583069},{"id":"https://openalex.org/C26022165","wikidata":"https://www.wikidata.org/wiki/Q8091","display_name":"Grammar","level":2,"score":0.47130000591278076},{"id":"https://openalex.org/C44291984","wikidata":"https://www.wikidata.org/wiki/Q1074173","display_name":"Question answering","level":2,"score":0.4138999879360199},{"id":"https://openalex.org/C2779843651","wikidata":"https://www.wikidata.org/wiki/Q7390335","display_name":"SIGNAL (programming language)","level":2,"score":0.3449000120162964},{"id":"https://openalex.org/C184337299","wikidata":"https://www.wikidata.org/wiki/Q1437428","display_name":"Semantics (computer science)","level":2,"score":0.3440999984741211},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.31630000472068787},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.2851000130176544},{"id":"https://openalex.org/C129792486","wikidata":"https://www.wikidata.org/wiki/Q1050419","display_name":"Language identification","level":3,"score":0.2849000096321106},{"id":"https://openalex.org/C28490314","wikidata":"https://www.wikidata.org/wiki/Q189436","display_name":"Speech recognition","level":1,"score":0.2840000092983246},{"id":"https://openalex.org/C34146451","wikidata":"https://www.wikidata.org/wiki/Q5048094","display_name":"Cascade","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C195324797","wikidata":"https://www.wikidata.org/wiki/Q33742","display_name":"Natural language","level":2,"score":0.27489998936653137},{"id":"https://openalex.org/C2776187449","wikidata":"https://www.wikidata.org/wiki/Q1513879","display_name":"Natural language generation","level":3,"score":0.2728999853134155},{"id":"https://openalex.org/C74672266","wikidata":"https://www.wikidata.org/wiki/Q815859","display_name":"Language acquisition","level":2,"score":0.2583000063896179},{"id":"https://openalex.org/C60048249","wikidata":"https://www.wikidata.org/wiki/Q37437","display_name":"Syntax","level":2,"score":0.2567000091075897}],"mesh":[],"locations_count":2,"locations":[{"id":"pmh:oai:arXiv.org:2602.12005","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2602.12005","pdf_url":"https://arxiv.org/pdf/2602.12005","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":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},{"id":"doi:10.48550/arxiv.2602.12005","doi":"https://doi.org/10.48550/arxiv.2602.12005","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2602.12005","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":"pmh:oai:arXiv.org:2602.12005","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2602.12005","pdf_url":"https://arxiv.org/pdf/2602.12005","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":"cc-by-nc-nd","license_id":"https://openalex.org/licenses/cc-by-nc-nd","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":null,"raw_type":"text"},"sustainable_development_goals":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7128782817.pdf","grobid_xml":"https://content.openalex.org/works/W7128782817.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W7213897969","https://openalex.org/W7166869576","https://openalex.org/W4226455589","https://openalex.org/W7166877953","https://openalex.org/W4415318184","https://openalex.org/W3113280695","https://openalex.org/W7166881382","https://openalex.org/W4402670143","https://openalex.org/W7128312581","https://openalex.org/W4385571795"],"abstract_inverted_index":{"Language":[0,33],"models":[1,232],"have":[2],"consistently":[3],"grown":[4],"to":[5,39,53,59,131,141,219,237,241],"compress":[6],"more":[7],"world":[8],"knowledge":[9,15],"into":[10,20],"their":[11,25],"parameters,":[12],"but":[13,166],"the":[14,29,50,57,73,109,119,172,178,187],"that":[16,99,158,185,213,230],"can":[17,81],"be":[18,161],"pretrained":[19],"them":[21],"is":[22,36,45,101,111,122],"upper-bounded":[23],"by":[24,48],"parameter":[26],"size.":[27],"Especially":[28],"capacity":[30],"of":[31,76,106,113,145,174],"Small":[32],"Models":[34],"(SLMs)":[35],"limited,":[37],"leading":[38],"factually":[40],"incorrect":[41],"generations.":[42],"This":[43,156,245],"problem":[44],"often":[46],"mitigated":[47],"giving":[49],"SLM":[51,80,224],"access":[52],"an":[54,79,223],"outside":[55],"source:":[56],"ability":[58],"query":[60],"a":[61,66,94,104,115,146,154,175,195,209,253,256],"larger":[62],"model,":[63],"documents,":[64],"or":[65,133,262],"database.":[67],"Under":[68],"this":[69,100,204],"setting,":[70],"we":[71,183,206],"study":[72],"fundamental":[74],"question":[75,105],"\\emph{which":[77,88],"tokens":[78,139,222,236],"and":[82,149,239,259,269],"should":[83,91,151,225],"learn}":[84],"during":[85],"pretraining,":[86],"versus":[87],"ones":[89],"it":[90,121],"delegate}":[92],"via":[93],"\\texttt{}":[95],"token.":[96],"We":[97],"find":[98],"not":[102,152],"simply":[103],"loss:":[107],"although":[108],"loss":[110,164,188,215],"predictive":[112],"whether":[114],"predicted":[116],"token":[117,176],"mismatches":[118],"ground-truth,":[120],"insufficient":[123],"for":[124,243],"identifying":[125],"which":[126,221,235],"predictions":[127],"would":[128],"actually":[129],"lead":[130],"factual":[132],"semantically":[134],"invalid":[135],"continuations.":[136],"Some":[137],"high-loss":[138],"correspond":[140],"\\emph{acceptable}":[142],"alternative":[143],"continuations":[144],"pretraining":[147,211],"document":[148],"therefore":[150],"trigger":[153],"\\texttt{}.":[155],"suggests":[157],"learnability":[159],"cannot":[160],"characterized":[162],"from":[163,194],"alone,":[165],"requires":[167],"additional":[168],"domain-specific":[169],"signals":[170,218],"about":[171],"role":[173],"in":[177,247,252],"sentence.":[179],"In":[180],"Wikipedia-like":[181],"domains,":[182],"show":[184],"augmenting":[186],"signal":[189],"with":[190,216,255],"lightweight":[191],"grammatical":[192],"information":[193],"spaCy":[196],"parser":[197],"substantially":[198],"improves":[199],"delegation":[200],"decisions.":[201],"Based":[202],"on":[203],"insight,":[205],"propose":[207],"LaCy,":[208],"novel":[210],"method":[212],"combines":[214],"factuality":[217],"decide":[220],"learn.":[226],"Our":[227],"experiments":[228],"demonstrate":[229],"LaCy":[231],"successfully":[233],"learn":[234],"predict":[238],"when":[240,250],"call":[242],"help.":[244],"results":[246],"higher":[248],"FactScores":[249],"generating":[251],"cascade":[254],"bigger":[257],"model":[258],"outperforms":[260],"Rho":[261],"LLM-judge":[263],"trained":[264],"SLMs,":[265],"while":[266],"being":[267],"simpler":[268],"cheaper.":[270]},"counts_by_year":[],"updated_date":"2026-10-28T10:33:24.094632","created_date":"2026-02-14T00:00:00"}