{"id":"https://openalex.org/W7213675608","doi":null,"title":"Compositional Reasoning in Language Models under Reinforcement Learning Post-Training","display_name":"Compositional Reasoning in Language Models under Reinforcement Learning Post-Training","publication_year":2026,"publication_date":"2026-09-16","ids":{"openalex":"https://openalex.org/W7213675608"},"language":"en","primary_location":{"id":"pmh:oai:arXiv.org:2609.19465","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.19465","pdf_url":"https://arxiv.org/pdf/2609.19465","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"},"volume":null,"type":"preprint","is_traditional":true,"indexed_in":["arxiv"],"open_access":{"is_oa":true,"oa_status":"green","oa_url":"https://arxiv.org/pdf/2609.19465","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5152179250","display_name":"Yu He","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"He, Yu","raw_affiliation_strings":["Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Stanford University","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5150972338","display_name":"Yingxi Li","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Li, Yingxi","raw_affiliation_strings":["Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Stanford University","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5152222921","display_name":"Yifei Wang","orcid":null,"observed_orcids":[]},"institutions":[{"id":"https://openalex.org/I1311688040","display_name":"Amazon (United States)","ror":"https://ror.org/04mv4n011","country_code":"US","type":"company","lineage":["https://openalex.org/I1311688040"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Wang, Yifei","raw_affiliation_strings":["Amazon AGI Labs"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Amazon AGI Labs","institution_ids":["https://openalex.org/I1311688040"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5050640377","display_name":"Ellen Vitercik","orcid":"https://orcid.org/0000-0003-4891-1367","observed_orcids":["https://orcid.org/0000-0003-4891-1367"]},"institutions":[{"id":"https://openalex.org/I97018004","display_name":"Stanford University","ror":"https://ror.org/00f54p054","country_code":"US","type":"education","lineage":["https://openalex.org/I97018004"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Vitercik, Ellen","raw_affiliation_strings":["Stanford University"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Stanford University","institution_ids":["https://openalex.org/I97018004"]}]}],"institutions":[],"countries_distinct_count":1,"institutions_distinct_count":2,"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":0.3684000074863434,"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":0.3684000074863434,"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/T11307","display_name":"Domain Adaptation and Few-Shot Learning","score":0.18930000066757202,"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":0.147599995136261,"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/compositional-reasoning","display_name":"compositional reasoning","score":1.0},{"id":"https://openalex.org/keywords/language-models","display_name":"language models","score":0.7639999985694885},{"id":"https://openalex.org/keywords/reinforcement-learning-post-training","display_name":"reinforcement learning post-training","score":0.9729999899864197},{"id":"https://openalex.org/keywords/skill-composition","display_name":"skill composition","score":0.6060000061988831},{"id":"https://openalex.org/keywords/compositional-generalization","display_name":"compositional generalization","score":0.7990000247955322},{"id":"https://openalex.org/keywords/length-extrapolation","display_name":"length extrapolation","score":0.9369999766349792},{"id":"https://openalex.org/keywords/transfer-learning","display_name":"transfer learning","score":0.4959999918937683}],"concepts":[{"id":"https://openalex.org/C121375916","wikidata":"https://www.wikidata.org/wiki/Q936559","display_name":"Principle of compositionality","level":2,"score":0.8614000082015991},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7217000126838684},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6620000004768372},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6396999955177307},{"id":"https://openalex.org/C97541855","wikidata":"https://www.wikidata.org/wiki/Q830687","display_name":"Reinforcement learning","level":2,"score":0.5698000192642212},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.49709999561309814},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.39340001344680786},{"id":"https://openalex.org/C150899416","wikidata":"https://www.wikidata.org/wiki/Q1820378","display_name":"Transfer of learning","level":2,"score":0.3702999949455261},{"id":"https://openalex.org/C45374587","wikidata":"https://www.wikidata.org/wiki/Q12525525","display_name":"Computation","level":2,"score":0.3626999855041504},{"id":"https://openalex.org/C2983448237","wikidata":"https://www.wikidata.org/wiki/Q1078276","display_name":"Language understanding","level":2,"score":0.3407000005245209},{"id":"https://openalex.org/C2777938197","wikidata":"https://www.wikidata.org/wiki/Q7834022","display_name":"Transfer of training","level":2,"score":0.328900009393692},{"id":"https://openalex.org/C168167062","wikidata":"https://www.wikidata.org/wiki/Q1117970","display_name":"Component (thermodynamics)","level":2,"score":0.3264999985694885},{"id":"https://openalex.org/C189950617","wikidata":"https://www.wikidata.org/wiki/Q937228","display_name":"Property (philosophy)","level":2,"score":0.31470000743865967},{"id":"https://openalex.org/C188147891","wikidata":"https://www.wikidata.org/wiki/Q147638","display_name":"Cognitive science","level":1,"score":0.2840999960899353},{"id":"https://openalex.org/C137293760","wikidata":"https://www.wikidata.org/wiki/Q3621696","display_name":"Language model","level":2,"score":0.2777999937534332},{"id":"https://openalex.org/C67203356","wikidata":"https://www.wikidata.org/wiki/Q1321905","display_name":"Reinforcement","level":2,"score":0.27070000767707825},{"id":"https://openalex.org/C132758656","wikidata":"https://www.wikidata.org/wiki/Q5307365","display_name":"Dreyfus model of skill acquisition","level":2,"score":0.27070000767707825},{"id":"https://openalex.org/C2778334786","wikidata":"https://www.wikidata.org/wiki/Q1586270","display_name":"Variation (astronomy)","level":2,"score":0.2671999931335449},{"id":"https://openalex.org/C2776502983","wikidata":"https://www.wikidata.org/wiki/Q690182","display_name":"Contrast (vision)","level":2,"score":0.2646999955177307},{"id":"https://openalex.org/C66024118","wikidata":"https://www.wikidata.org/wiki/Q1122506","display_name":"Computational model","level":2,"score":0.26409998536109924},{"id":"https://openalex.org/C97364631","wikidata":"https://www.wikidata.org/wiki/Q484284","display_name":"Deductive reasoning","level":2,"score":0.259799987077713},{"id":"https://openalex.org/C51632099","wikidata":"https://www.wikidata.org/wiki/Q3985153","display_name":"Training set","level":2,"score":0.2578999996185303},{"id":"https://openalex.org/C193221554","wikidata":"https://www.wikidata.org/wiki/Q5153664","display_name":"Commonsense reasoning","level":2,"score":0.2554999887943268}],"mesh":[],"locations_count":1,"locations":[{"id":"pmh:oai:arXiv.org:2609.19465","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.19465","pdf_url":"https://arxiv.org/pdf/2609.19465","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"}],"best_oa_location":{"id":"pmh:oai:arXiv.org:2609.19465","doi":null,"is_oa":true,"landing_page_url":"https://arxiv.org/abs/2609.19465","pdf_url":"https://arxiv.org/pdf/2609.19465","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"},"sustainable_development_goals":[],"study_designs":[],"awards":[{"id":"https://openalex.org/G1391553139","display_name":"CAREER: Machine Learning for Discrete Optimization","funder_award_id":"2338226","funder_id":"https://openalex.org/F4320306076","funder_display_name":"National Science Foundation"}],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W7213675608.pdf","grobid_xml":"https://content.openalex.org/works/W7213675608.grobid-xml"},"referenced_works_count":11,"referenced_works":["https://openalex.org/W3104739822","https://openalex.org/W4391673329","https://openalex.org/W4401042708","https://openalex.org/W4403883066","https://openalex.org/W4414281281","https://openalex.org/W4414856006","https://openalex.org/W4414939355","https://openalex.org/W4416048576","https://openalex.org/W7083820855","https://openalex.org/W7128804817","https://openalex.org/W7133191918"],"related_works":["https://openalex.org/W4416967228","https://openalex.org/W7165006695","https://openalex.org/W4416048576","https://openalex.org/W4385571345","https://openalex.org/W4407309390","https://openalex.org/W7083820855","https://openalex.org/W4414586223","https://openalex.org/W7167814253","https://openalex.org/W7128491810","https://openalex.org/W4367692257"],"abstract_inverted_index":{"Compositional":[0],"reasoning":[1,36,46],"is":[2,11],"critical":[3],"for":[4,114],"real-world":[5,142],"problem":[6],"solving:":[7],"since":[8],"training":[9,92,102],"data":[10],"necessarily":[12],"limited,":[13],"models":[14,40],"must":[15],"generalize":[16],"by":[17],"composing":[18],"learned":[19],"skills":[20],"in":[21],"new":[22],"ways.":[23],"While":[24],"post-training":[25],"methods":[26],"such":[27],"as":[28],"reinforcement":[29],"learning":[30],"(RL)":[31],"have":[32],"substantially":[33],"improved":[34],"the":[35,149],"abilities":[37],"of":[38,63],"language":[39],"(LMs),":[41],"their":[42],"effects":[43],"on":[44,141],"compositional":[45,58,83,120],"remain":[47],"less":[48],"well":[49],"understood.":[50],"We":[51,85,110],"propose":[52],"a":[53,87,138],"dependency-graph":[54],"framework":[55,72],"to":[56,97,107,130,154],"formalize":[57],"reasoning,":[59],"yielding":[60],"three":[61],"levels":[62],"compositionality":[64],"with":[65,73],"increasing":[66],"complexity.":[67],"Empirically,":[68],"we":[69,136],"instantiate":[70],"this":[71,115],"data-structure":[74],"tasks,":[75,99],"which":[76],"provide":[77,111],"deterministic":[78],"reward":[79],"computation":[80],"and":[81,117,128],"clear":[82],"structure.":[84],"find":[86],"consistent":[88],"decomposed-to-composed":[89,150],"asymmetry:":[90],"decomposed-skill":[91],"does":[93],"not":[94],"reliably":[95],"transfer":[96,129],"composed":[98],"whereas":[100],"composed-task":[101],"transfers":[103],"more":[104],"readily":[105],"back":[106],"decomposed":[108],"tasks.":[109],"theoretical":[112],"explanation":[113],"asymmetry,":[116],"further":[118],"evaluate":[119],"generalization":[121],"under":[122],"length":[123],"extrapolation,":[124],"structural":[125],"distribution":[126],"shift,":[127],"tasks":[131],"requiring":[132],"unseen":[133],"skills.":[134],"Finally,":[135],"present":[137],"pilot":[139],"study":[140],"tool-calling":[143],"benchmarks,":[144],"showing":[145],"preliminary":[146],"evidence":[147],"that":[148],"asymmetry":[151],"can":[152],"extend":[153],"practical":[155],"settings.":[156]},"counts_by_year":[],"updated_date":"2026-11-07T11:25:07.637180","created_date":"2026-09-20T00:00:00"}