{"id":"https://openalex.org/W4412944795","doi":"https://doi.org/10.18653/v1/2025.acl-long.1104","title":"S2R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning","display_name":"S2R: Teaching LLMs to Self-verify and Self-correct via Reinforcement Learning","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412944795","doi":"https://doi.org/10.18653/v1/2025.acl-long.1104"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.acl-long.1104","doi":"https://doi.org/10.18653/v1/2025.acl-long.1104","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1104","pdf_url":"https://aclanthology.org/2025.acl-long.1104.pdf","source":{"id":"https://openalex.org/S4393915740","display_name":"Proceedings of the conference - Association for Computational Linguistics. Meeting","issn_l":"0736-587X","issn":["0736-587X"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"volume":null,"type":"conference-paper","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"bronze","oa_url":"https://aclanthology.org/2025.acl-long.1104.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Ruotian Ma","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Ruotian Ma","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Peisong Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I99065089","display_name":"Tsinghua University","ror":"https://ror.org/03cve4549","country_code":"CN","type":"education","lineage":["https://openalex.org/I99065089"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Peisong Wang","raw_affiliation_strings":["Tsinghua University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tsinghua University","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100384275","display_name":"Cheng Liu","orcid":"https://orcid.org/0000-0001-8967-1174","observed_orcids":["https://orcid.org/0000-0001-8967-1174"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Cheng Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5101688566","display_name":"Xingyan Liu","orcid":"https://orcid.org/0000-0001-5117-0606","observed_orcids":["https://orcid.org/0000-0001-5117-0606"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Xingyan Liu","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100434197","display_name":"Jiaqi Chen","orcid":"https://orcid.org/0000-0002-4069-193X","observed_orcids":["https://orcid.org/0000-0002-4069-193X"]},"institutions":[{"id":"https://openalex.org/I889458895","display_name":"University of Hong Kong","ror":"https://ror.org/02zhqgq86","country_code":"HK","type":"education","lineage":["https://openalex.org/I889458895"]}],"countries":["HK"],"is_corresponding":false,"raw_author_name":"Jiaqi Chen","raw_affiliation_strings":["The University of Hong Kong"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The University of Hong Kong","institution_ids":["https://openalex.org/I889458895"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5082907437","display_name":"Bang Zhang","orcid":null,"observed_orcids":[]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Bang Zhang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":null,"display_name":"Xin Zhou","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I24943067","display_name":"Fudan University","ror":"https://ror.org/013q1eq08","country_code":"CN","type":"education","lineage":["https://openalex.org/I24943067"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Xin Zhou","raw_affiliation_strings":["Fudan University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Fudan University","institution_ids":["https://openalex.org/I24943067"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5100721790","display_name":"Nan Du","orcid":"https://orcid.org/0000-0003-2855-7452","observed_orcids":["https://orcid.org/0000-0003-2855-7452","https://orcid.org/0000-0003-1762-7275"]},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Nan Du","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5100648221","display_name":"Jia Li","orcid":"https://orcid.org/0000-0002-1824-9379","observed_orcids":["https://orcid.org/0000-0002-1824-9379"]},"institutions":[{"id":"https://openalex.org/I4405261251","display_name":"The Hong Kong University of Science and Technology (Guangzhou)","ror":"https://ror.org/050h0vm43","country_code":"CN","type":"education","lineage":["https://openalex.org/I4405261251"]}],"countries":["CN"],"is_corresponding":false,"raw_author_name":"Jia Li","raw_affiliation_strings":["The Hong Kong University of Science and Technology (Guangzhou )"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"The Hong Kong University of Science and Technology (Guangzhou )","institution_ids":["https://openalex.org/I4405261251"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":4.7482,"has_fulltext":true,"cited_by_count":5,"citation_normalized_percentile":{"value":0.95490149,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":90,"max":98},"biblio":{"volume":null,"issue":null,"first_page":"22632","last_page":"22654"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.4943999946117401,"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":0.4943999946117401,"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":0.2815000116825104,"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":0.037300001829862595,"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/large-language-models","display_name":"large language models","score":0.9269999861717224},{"id":"https://openalex.org/keywords/self-verification","display_name":"self-verification","score":0.9279999732971191},{"id":"https://openalex.org/keywords/self-correction","display_name":"self-correction","score":0.9850000143051147},{"id":"https://openalex.org/keywords/reinforcement-learning","display_name":"reinforcement learning","score":0.8849999904632568},{"id":"https://openalex.org/keywords/personal-development","display_name":"human development","score":0.4180000126361847}],"concepts":[{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.6829149723052979},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6248979568481445},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.5328382253646851},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.4097749590873718},{"id":"https://openalex.org/C15744967","wikidata":"https://www.wikidata.org/wiki/Q9418","display_name":"Psychology","level":0,"score":0.17672225832939148},{"id":"https://openalex.org/C77805123","wikidata":"https://www.wikidata.org/wiki/Q161272","display_name":"Social psychology","level":1,"score":0.12886488437652588}],"mesh":[],"locations_count":3,"locations":[{"id":"doi:10.18653/v1/2025.acl-long.1104","doi":"https://doi.org/10.18653/v1/2025.acl-long.1104","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1104","pdf_url":"https://aclanthology.org/2025.acl-long.1104.pdf","source":{"id":"https://openalex.org/S4393915740","display_name":"Proceedings of the conference - Association for Computational Linguistics. Meeting","issn_l":"0736-587X","issn":["0736-587X"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:2502.12853","doi":null,"is_oa":true,"landing_page_url":"http://arxiv.org/abs/2502.12853","pdf_url":"https://arxiv.org/pdf/2502.12853","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.2502.12853","doi":"https://doi.org/10.48550/arxiv.2502.12853","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.2502.12853","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/v1/2025.acl-long.1104","doi":"https://doi.org/10.18653/v1/2025.acl-long.1104","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.acl-long.1104","pdf_url":"https://aclanthology.org/2025.acl-long.1104.pdf","source":{"id":"https://openalex.org/S4393915740","display_name":"Proceedings of the conference - Association for Computational Linguistics. Meeting","issn_l":"0736-587X","issn":["0736-587X"],"is_oa":false,"is_in_doaj":false,"is_core":false,"listed_in":["norway-1"],"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"host_organization_lineage_names":[],"type":"conference"},"license":null,"license_id":null,"version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)","raw_type":"proceedings-article"},"sustainable_development_goals":[],"sustainable_development_goals_aurora":[],"study_designs":[],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412944795.pdf","grobid_xml":"https://content.openalex.org/works/W4412944795.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W4407806944","https://openalex.org/W4414896360","https://openalex.org/W4416456107","https://openalex.org/W7202360075","https://openalex.org/W4416149163","https://openalex.org/W4417351618","https://openalex.org/W4403885103","https://openalex.org/W7127704939","https://openalex.org/W4402669701","https://openalex.org/W6929297702"],"abstract_inverted_index":{"Recent":[0],"studies":[1],"have":[2],"demonstrated":[3],"the":[4,32,128,154],"effectiveness":[5,155],"of":[6,35,122,130,156],"LLM":[7,52],"test-time":[8],"scaling.However,":[9],"existing":[10],"approaches":[11],"to":[12,30,57,113],"incentivize":[13],"LLMs'":[14,136],"deep":[15,137],"thinking":[16,33],"abilities":[17,34],"generally":[18],"require":[19],"large-scale":[20],"data":[21],"or":[22],"significant":[23],"training":[24],"efforts.Meanwhile,":[25],"it":[26],"remains":[27],"unclear":[28],"how":[29],"improve":[31],"less":[36],"powerful":[37],"base":[38,145],"models.In":[39],"this":[40],"work,":[41],"we":[42,63],"introduce":[43],"S":[44,157],"2":[45,158],"R,":[46],"an":[47,108,119],"efficient":[48],"framework":[49],"that":[50],"enhances":[51],"reasoning":[53],"by":[54,88],"teaching":[55],"models":[56,116,146],"self-verify":[58],"and":[59,70,81,90,140,150],"self-correct":[60],"during":[61],"inference.Specifically,":[62],"first":[64],"initialize":[65],"LLMs":[66],"with":[67,94,101],"iterative":[68],"self-verification":[69,80],"self-correction":[71,82],"behaviors":[72],"through":[73],"supervised":[74],"fine-tuning":[75],"on":[76,118,134,143],"carefully":[77],"curated":[78],"data.The":[79],"skills":[83],"are":[84],"then":[85],"further":[86],"strengthened":[87],"outcome-level":[89],"process-level":[91],"reinforcement":[92],"learning":[93],"minimized":[95],"resource":[96],"requirements.Our":[97],"results":[98],"demonstrate":[99],"that,":[100],"only":[102],"3.1k":[103],"behavior":[104],"initialization":[105],"samples,":[106],"Qwen2.5-math-7Bachieves":[107],"accuracy":[109],"improvement":[110],"from":[111],"51.0%":[112],"81.6%,":[114],"outperforming":[115],"trained":[117],"equivalent":[120],"amount":[121],"long-CoT":[123],"distilled":[124],"data.We":[125],"also":[126],"discuss":[127],"effect":[129],"different":[131],"RL":[132],"strategies":[133],"enhancing":[135],"reasoning.Extensive":[138],"experiments":[139],"analysis":[141],"based":[142],"three":[144],"across":[147],"both":[148],"in-domain":[149],"out-of-domain":[151],"benchmarks":[152],"validate":[153],"R":[159],"1":[160],".":[161]},"counts_by_year":[{"year":2026,"cited_by_count":4},{"year":2025,"cited_by_count":1}],"updated_date":"2026-10-29T09:56:32.521200","created_date":"2025-10-10T00:00:00"}