{"id":"https://openalex.org/W4412888475","doi":"https://doi.org/10.18653/v1/2025.findings-acl.493","title":"Implicit Reasoning in Transformers is Reasoning through Shortcuts","display_name":"Implicit Reasoning in Transformers is Reasoning through Shortcuts","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4412888475","doi":"https://doi.org/10.18653/v1/2025.findings-acl.493"},"language":"en","primary_location":{"id":"doi:10.18653/v1/2025.findings-acl.493","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.493","pdf_url":"https://aclanthology.org/2025.findings-acl.493.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 2025","raw_type":"proceedings-article"},"type":"conference-paper","indexed_in":["crossref"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://aclanthology.org/2025.findings-acl.493.pdf","any_repository_has_fulltext":null},"authorships":[{"author_position":"first","author":{"id":null,"display_name":"Tianhe Lin","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Tianhe Lin","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5102637896","display_name":"Jian Xie","orcid":null},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Jian Xie","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"middle","author":{"id":"https://openalex.org/A5067972035","display_name":"Siyu Yuan","orcid":"https://orcid.org/0000-0001-8161-6429"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Siyu Yuan","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]},{"author_position":"last","author":{"id":"https://openalex.org/A5046589466","display_name":"Deqing Yang","orcid":"https://orcid.org/0000-0003-1958-7182"},"institutions":[],"countries":[],"is_corresponding":false,"raw_author_name":"Deqing Yang","raw_affiliation_strings":[],"raw_orcid":null,"affiliations":[]}],"institutions":[],"countries_distinct_count":0,"institutions_distinct_count":0,"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":"9470","last_page":"9487"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.5803999900817871,"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/T11010","display_name":"Logic, Reasoning, and Knowledge","score":0.5803999900817871,"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/T10215","display_name":"Semantic Web and Ontologies","score":0.5116000175476074,"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/T12002","display_name":"Computability, Logic, AI Algorithms","score":0.4772999882698059,"subfield":{"id":"https://openalex.org/subfields/1703","display_name":"Computational Theory and Mathematics"},"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/transformer","display_name":"Transformer","score":0.6587234735488892},{"id":"https://openalex.org/keywords/computer-science","display_name":"Computer science","score":0.6513100266456604},{"id":"https://openalex.org/keywords/model-based-reasoning","display_name":"Model-based reasoning","score":0.5669094324111938},{"id":"https://openalex.org/keywords/opportunistic-reasoning","display_name":"Opportunistic reasoning","score":0.4734097421169281},{"id":"https://openalex.org/keywords/deductive-reasoning","display_name":"Deductive reasoning","score":0.4595851004123688},{"id":"https://openalex.org/keywords/case-based-reasoning","display_name":"Case-based reasoning","score":0.45824262499809265},{"id":"https://openalex.org/keywords/reasoning-system","display_name":"Reasoning system","score":0.435604453086853},{"id":"https://openalex.org/keywords/analytic-reasoning","display_name":"Analytic reasoning","score":0.43441325426101685},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.29673343896865845},{"id":"https://openalex.org/keywords/knowledge-representation-and-reasoning","display_name":"Knowledge representation and reasoning","score":0.14138451218605042},{"id":"https://openalex.org/keywords/engineering","display_name":"Engineering","score":0.1113872230052948},{"id":"https://openalex.org/keywords/electrical-engineering","display_name":"Electrical engineering","score":0.08633533120155334}],"concepts":[{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6587234735488892},{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.6513100266456604},{"id":"https://openalex.org/C37335422","wikidata":"https://www.wikidata.org/wiki/Q6888134","display_name":"Model-based reasoning","level":3,"score":0.5669094324111938},{"id":"https://openalex.org/C86827895","wikidata":"https://www.wikidata.org/wiki/Q7098582","display_name":"Opportunistic reasoning","level":4,"score":0.4734097421169281},{"id":"https://openalex.org/C97364631","wikidata":"https://www.wikidata.org/wiki/Q484284","display_name":"Deductive reasoning","level":2,"score":0.4595851004123688},{"id":"https://openalex.org/C20162079","wikidata":"https://www.wikidata.org/wiki/Q1151406","display_name":"Case-based reasoning","level":2,"score":0.45824262499809265},{"id":"https://openalex.org/C89288958","wikidata":"https://www.wikidata.org/wiki/Q7301504","display_name":"Reasoning system","level":2,"score":0.435604453086853},{"id":"https://openalex.org/C103057564","wikidata":"https://www.wikidata.org/wiki/Q4751139","display_name":"Analytic reasoning","level":3,"score":0.43441325426101685},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.29673343896865845},{"id":"https://openalex.org/C161301231","wikidata":"https://www.wikidata.org/wiki/Q3478658","display_name":"Knowledge representation and reasoning","level":2,"score":0.14138451218605042},{"id":"https://openalex.org/C127413603","wikidata":"https://www.wikidata.org/wiki/Q11023","display_name":"Engineering","level":0,"score":0.1113872230052948},{"id":"https://openalex.org/C119599485","wikidata":"https://www.wikidata.org/wiki/Q43035","display_name":"Electrical engineering","level":1,"score":0.08633533120155334},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0}],"mesh":[],"locations_count":1,"locations":[{"id":"doi:10.18653/v1/2025.findings-acl.493","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.493","pdf_url":"https://aclanthology.org/2025.findings-acl.493.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 2025","raw_type":"proceedings-article"}],"best_oa_location":{"id":"doi:10.18653/v1/2025.findings-acl.493","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/2025.findings-acl.493","pdf_url":"https://aclanthology.org/2025.findings-acl.493.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 2025","raw_type":"proceedings-article"},"sustainable_development_goals":[],"awards":[],"funders":[{"id":"https://openalex.org/F4320306076","display_name":"National Science Foundation","ror":"https://ror.org/021nxhr62"},{"id":"https://openalex.org/F4320321851","display_name":"Fudan University","ror":"https://ror.org/013q1eq08"},{"id":"https://openalex.org/F4320335577","display_name":"Major Research Plan","ror":null}],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4412888475.pdf","grobid_xml":"https://content.openalex.org/works/W4412888475.grobid-xml"},"referenced_works_count":0,"referenced_works":[],"related_works":["https://openalex.org/W2148216026","https://openalex.org/W2351118185","https://openalex.org/W2188850991","https://openalex.org/W2077968118","https://openalex.org/W2199920951","https://openalex.org/W2154137762","https://openalex.org/W776051173","https://openalex.org/W2183503013","https://openalex.org/W2158160374","https://openalex.org/W3149557148"],"abstract_inverted_index":{"Test-time":[0],"compute":[1],"is":[2,39,143],"emerging":[3,124],"as":[4,16,26,28],"a":[5,68,133],"new":[6],"paradigm":[7],"for":[8],"enhancing":[9],"language":[10,81,149,154],"models'":[11],"complex":[12],"multi-step":[13,70,87],"reasoning":[14,33,38,50,58,72,85,97,122,158],"capabilities,":[15],"demonstrated":[17],"by":[18],"the":[19,48,56,176],"success":[20],"of":[21],"OpenAI's":[22],"o1":[23],"and":[24,74,98,105,136],"o3,":[25],"well":[27],"DeepSeek's":[29],"R1.Compared":[30],"to":[31,53,78,131,138],"explicit":[32],"in":[34,55,86,102,146],"test-time":[35],"compute,":[36],"implicit":[37,57,84,109,121,157],"more":[40],"inference-efficient,":[41],"requiring":[42],"fewer":[43],"generated":[44],"tokens.However,":[45],"why":[46],"does":[47],"advanced":[49],"capability":[51,112],"fail":[52,137],"emerge":[54],"style?In":[59],"this":[60,111,141],"work,":[61],"we":[62],"train":[63],"GPT-2":[64],"from":[65,125],"scratch":[66],"on":[67,117,127,165,175],"curated":[69],"mathematical":[71],"dataset":[73],"conduct":[75],"analytical":[76],"experiments":[77],"investigate":[79],"how":[80],"models":[82,93,155],"perform":[83,95],"tasks.Our":[88],"findings":[89,151],"reveal:":[90],"1)":[91],"Language":[92],"can":[94],"step-by-step":[96],"achieve":[99],"high":[100],"accuracy":[101],"both":[103],"in-domain":[104],"out-of-domain":[106],"tests":[107],"via":[108],"reasoning.However,":[110],"only":[113],"emerges":[114],"when":[115],"trained":[116],"fixed-pattern":[118],"data.2)":[119],"Conversely,":[120],"abilities":[123],"training":[126],"unfixed-pattern":[128],"data":[129],"tend":[130],"overfit":[132],"specific":[134],"pattern":[135],"generalize":[139],"further.Notably,":[140],"limitation":[142],"also":[144],"observed":[145],"state-of-the-art":[147],"large":[148],"models.These":[150],"suggest":[152],"that":[153],"acquire":[156],"through":[159],"shortcut":[160],"learning,":[161],"enabling":[162],"strong":[163],"performance":[164],"tasks":[166],"with":[167],"similar":[168],"patterns":[169],"while":[170],"lacking":[171],"generalization.Resources":[172],"are":[173],"available":[174],"GitHub.":[177]},"counts_by_year":[],"updated_date":"2026-07-14T23:27:15.235271","created_date":"2025-10-10T00:00:00"}