{"id":"https://openalex.org/W3112784227","doi":"https://doi.org/10.18653/v1/2021.findings-acl.74","title":"As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages","display_name":"As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages","publication_year":2021,"publication_date":"2021-01-01","ids":{"openalex":"https://openalex.org/W3112784227","doi":"https://doi.org/10.18653/v1/2021.findings-acl.74","mag":"3112784227"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2021.findings-acl.74","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2021.findings-acl.74","pdf_url":"https://aclanthology.org/2021.findings-acl.74.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-IJCNLP 2021","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["arxiv","crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2021.findings-acl.74.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Wietse de Vries","orcid":null},"institutions":[{"id":"https://openalex.org/I169381384","display_name":"University of Groningen","ror":"https://ror.org/012p63287","country_code":"NL","type":"education","lineage":["https://openalex.org/I169381384"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Wietse de Vries","raw_affiliation_strings":["University of Groningen The Netherlands"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Groningen The Netherlands","institution_ids":["https://openalex.org/I169381384"]}]},{"author_position":"last","author":{"id":null,"display_name":"Malvina Nissim","orcid":null},"institutions":[{"id":"https://openalex.org/I169381384","display_name":"University of Groningen","ror":"https://ror.org/012p63287","country_code":"NL","type":"education","lineage":["https://openalex.org/I169381384"]}],"countries":["NL"],"is_corresponding":false,"raw_author_name":"Malvina Nissim","raw_affiliation_strings":["University of Groningen The Netherlands"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Groningen The Netherlands","institution_ids":["https://openalex.org/I169381384"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":1,"corresponding_author_ids":[],"corresponding_institution_ids":["https://openalex.org/I169381384"],"apc_list":null,"apc_paid":null,"fwci":4.9828,"has_fulltext":true,"cited_by_count":43,"citation_normalized_percentile":{"value":0.96280983,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":99},"biblio":{"volume":null,"issue":null,"first_page":"836","last_page":"846"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9998000264167786,"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":0.9998000264167786,"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":0.9968000054359436,"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/T13650","display_name":"Computational Physics and Python Applications","score":0.9646999835968018,"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/generative-grammar","display_name":"Generative grammar","score":0.5170000195503235},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.4941999912261963},{"id":"https://openalex.org/keywords/embedding","display_name":"Embedding","score":0.46639999747276306},{"id":"https://openalex.org/keywords/language-model","display_name":"Language model","score":0.44279998540878296},{"id":"https://openalex.org/keywords/retraining","display_name":"Retraining","score":0.42980000376701355},{"id":"https://openalex.org/keywords/lexical-item","display_name":"Lexical item","score":0.39169999957084656},{"id":"https://openalex.org/keywords/generative-model","display_name":"Generative model","score":0.367000013589859}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7067000269889832},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.6161999702453613},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6152999997138977},{"id":"https://openalex.org/C39890363","wikidata":"https://www.wikidata.org/wiki/Q36108","display_name":"Generative grammar","level":2,"score":0.5170000195503235},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.4941999912261963},{"id":"https://openalex.org/C41608201","wikidata":"https://www.wikidata.org/wiki/Q980509","display_name":"Embedding","level":2,"score":0.46639999747276306},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.44279998540878296},{"id":"https://openalex.org/C2778712577","wikidata":"https://www.wikidata.org/wiki/Q3505966","display_name":"Retraining","level":2,"score":0.42980000376701355},{"id":"https://openalex.org/C126706616","wikidata":"https://www.wikidata.org/wiki/Q2944660","display_name":"Lexical item","level":2,"score":0.39169999957084656},{"id":"https://openalex.org/C167966045","wikidata":"https://www.wikidata.org/wiki/Q5532625","display_name":"Generative model","level":3,"score":0.367000013589859},{"id":"https://openalex.org/C139807058","wikidata":"https://www.wikidata.org/wiki/Q352374","display_name":"Adaptation (eye)","level":2,"score":0.358599990606308},{"id":"https://openalex.org/C2778755073","wikidata":"https://www.wikidata.org/wiki/Q10858537","display_name":"Scale (ratio)","level":2,"score":0.3190999925136566},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.3183000087738037},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.30149999260902405},{"id":"https://openalex.org/C75778745","wikidata":"https://www.wikidata.org/wiki/Q342626","display_name":"Lag","level":2,"score":0.2736999988555908},{"id":"https://openalex.org/C155092808","wikidata":"https://www.wikidata.org/wiki/Q182557","display_name":"Computational linguistics","level":2,"score":0.2703000009059906},{"id":"https://openalex.org/C53893814","wikidata":"https://www.wikidata.org/wiki/Q7378909","display_name":"Rule-based machine translation","level":2,"score":0.26600000262260437}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.18653/v1/2021.findings-acl.74","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2021.findings-acl.74","pdf_url":"https://aclanthology.org/2021.findings-acl.74.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-IJCNLP 2021","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2012.05628","is_oa":true,"landing_page_url":"http://arxiv.org/abs/2012.05628","pdf_url":"https://arxiv.org/pdf/2012.05628","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,"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":"pmh:oai:pure.rug.nl:openaire/3c3049df-e64f-413e-aad2-74ce1e89ce42","is_oa":true,"landing_page_url":"https://research.rug.nl/en/publications/3c3049df-e64f-413e-aad2-74ce1e89ce42","pdf_url":null,"source":{"id":"https://openalex.org/S4306400420","display_name":"University of Groningen research database (University of Groningen / Centre for Information Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I169381384","host_organization_name":"University of Groningen","host_organization_lineage":["https://openalex.org/I169381384"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"de Vries, W & Nissim, M 2021 'As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages' ArXiv, arXiv. https://doi.org/10.18653/v1/2021.findings-acl.74","raw_type":"info:eu-repo/semantics/preprint"},{"id":"pmh:oai:pure.rug.nl:publications/3c3049df-e64f-413e-aad2-74ce1e89ce42","is_oa":true,"landing_page_url":"https://hdl.handle.net/11370/3c3049df-e64f-413e-aad2-74ce1e89ce42","pdf_url":null,"source":{"id":"https://openalex.org/S4306400420","display_name":"University of Groningen research database (University of Groningen / Centre for Information Technology)","issn_l":null,"issn":null,"is_oa":false,"is_in_doaj":false,"is_core":false,"host_organization":"https://openalex.org/I169381384","host_organization_name":"University of Groningen","host_organization_lineage":["https://openalex.org/I169381384"],"host_organization_lineage_names":[],"type":"repository"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"submittedVersion","is_accepted":false,"is_published":false,"raw_source_name":"de Vries, W & Nissim, M 2021 'As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages' ArXiv, arXiv. https://doi.org/10.18653/v1/2021.findings-acl.74","raw_type":"info:eu-repo/semantics/preprint"}],"best_oa_location":{"id":"doi:10.18653/v1/2021.findings-acl.74","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2021.findings-acl.74","pdf_url":"https://aclanthology.org/2021.findings-acl.74.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-IJCNLP 2021","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320320933","display_name":"Rijksuniversiteit Groningen","ror":"https://ror.org/012p63287"},{"id":"https://openalex.org/F4320321800","display_name":"Nederlandse Organisatie voor Wetenschappelijk Onderzoek","ror":"https://ror.org/04jsz6e67"},{"id":"https://openalex.org/F4320337373","display_name":"Center for Information Technology","ror":"https://ror.org/03jh5a977"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W3112784227.pdf","grobid_xml":"https://content.openalex.org/works/W3112784227.grobid-xml"},"referenced_works_count":1,"referenced_works":["https://openalex.org/W6674159460"],"related_works":[],"abstract_inverted_index":{"Large":[0],"generative":[1],"language":[2],"models":[3,36,120],"have":[4],"been":[5],"very":[6],"successful":[7],"for":[8,68],"English,":[9],"but":[10],"other":[11],"languages":[12],"lag":[13],"behind,":[14],"in":[15,129],"part":[16],"due":[17],"to":[18,37,48,94],"data":[19],"and":[20,50,70,107,131],"computational":[21],"limitations.":[22],"We":[23],"propose":[24],"a":[25,62,154],"method":[26,101],"that":[27,72,113],"may":[28],"overcome":[29],"these":[30,136],"problems":[31],"by":[32,52,86,116,143,153],"adapting":[33],"existing":[34],"pre-trained":[35],"new":[38],"languages.":[39],"Specifically,":[40],"we":[41,64,82],"describe":[42],"the":[43,58,76,95,103],"adaptation":[44,112],"of":[45,91,105],"English":[46,78,118],"GPT-2":[47,92,96,119,155],"Italian":[49,69,130],"Dutch":[51,71],"retraining":[53],"lexical":[54,66,79,89,123],"embeddings":[55,67,90,124],"without":[56],"tuning":[57],"Transformer":[59],"layers.":[60],"As":[61],"result,":[63],"obtain":[65],"are":[73,138,146],"aligned":[74],"with":[75,121,150],"original":[77],"embeddings.":[80],"Additionally,":[81],"scale":[83],"up":[84],"complexity":[85],"transforming":[87],"relearned":[88,122],"small":[93],"medium":[97],"embedding":[98],"space.":[99],"This":[100],"minimises":[102],"amount":[104],"training":[106],"prevents":[108],"losing":[109],"information":[110],"during":[111],"was":[114],"learned":[115],"GPT-2.":[117],"can":[125],"generate":[126],"realistic":[127],"sentences":[128,137,151],"Dutch.":[132],"Though":[133],"on":[134,148],"average":[135],"still":[139],"identifiable":[140],"as":[141],"artificial":[142],"humans,":[144],"they":[145],"assessed":[147],"par":[149],"generated":[152],"model":[156],"fully":[157],"trained":[158],"from":[159],"scratch.":[160]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":6},{"year":2024,"cited_by_count":9},{"year":2023,"cited_by_count":14},{"year":2022,"cited_by_count":4},{"year":2021,"cited_by_count":7}],"updated_date":"2026-08-05T07:39:15.569665","created_date":"2020-12-21T00:00:00"}