{"id":"https://openalex.org/W4407355147","doi":"https://doi.org/10.1162/tacl_a_00733","title":"Learning Syntax Without Planting Trees: Understanding Hierarchical Generalization in Transformers","display_name":"Learning Syntax Without Planting Trees: Understanding Hierarchical Generalization in Transformers","publication_year":2025,"publication_date":"2025-01-01","ids":{"openalex":"https://openalex.org/W4407355147","doi":"https://doi.org/10.1162/tacl_a_00733"},"language":"en","primary_location":{"id":"doi:10.1162/tacl_a_00733","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00733","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00733/2501791/tacl_a_00733.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"type":"article","indexed_in":["crossref","doaj"],"open_access":{"is_oa":true,"oa_status":"diamond","oa_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00733/2501791/tacl_a_00733.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5088355901","display_name":"Kabir Ahuja","orcid":null},"institutions":[{"id":"https://openalex.org/I201448701","display_name":"University of Washington","ror":"https://ror.org/00cvxb145","country_code":"US","type":"education","lineage":["https://openalex.org/I201448701"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Kabir Ahuja","raw_affiliation_strings":["University of Washington, USA. kahuja@cs.washington.edu"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Washington, USA. kahuja@cs.washington.edu","institution_ids":["https://openalex.org/I201448701"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5012491878","display_name":"Vidhisha Balachandran","orcid":"https://orcid.org/0009-0009-0465-0098"},"institutions":[{"id":"https://openalex.org/I1290206253","display_name":"Microsoft (United States)","ror":"https://ror.org/00d0nc645","country_code":"US","type":"company","lineage":["https://openalex.org/I1290206253"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Vidhisha Balachandran","raw_affiliation_strings":["Microsoft Research, USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research, USA","institution_ids":["https://openalex.org/I1290206253"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5014668625","display_name":"Madhur Panwar","orcid":null},"institutions":[{"id":"https://openalex.org/I5124864","display_name":"\u00c9cole Polytechnique F\u00e9d\u00e9rale de Lausanne","ror":"https://ror.org/02s376052","country_code":"CH","type":"education","lineage":["https://openalex.org/I2799323385","https://openalex.org/I5124864"]}],"countries":["CH"],"is_corresponding":false,"raw_author_name":"Madhur Panwar","raw_affiliation_strings":["EPFL, Switzerland"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"EPFL, Switzerland","institution_ids":["https://openalex.org/I5124864"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5051747323","display_name":"Tianxing He","orcid":"https://orcid.org/0009-0008-6383-0307"},"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":"Tianxing He","raw_affiliation_strings":["Tsinghua University, China"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Tsinghua University, China","institution_ids":["https://openalex.org/I99065089"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088517824","display_name":"Noah A. Smith","orcid":"https://orcid.org/0000-0002-2310-6380"},"institutions":[{"id":"https://openalex.org/I201448701","display_name":"University of Washington","ror":"https://ror.org/00cvxb145","country_code":"US","type":"education","lineage":["https://openalex.org/I201448701"]},{"id":"https://openalex.org/I4210140341","display_name":"Allen Institute","ror":"https://ror.org/03cpe7c52","country_code":"US","type":"nonprofit","lineage":["https://openalex.org/I4210140341"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Noah A. Smith","raw_affiliation_strings":["Allen Institute for AI, USA","University of Washington, USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Allen Institute for AI, USA","institution_ids":["https://openalex.org/I4210140341"]},{"raw_affiliation_string":"University of Washington, USA","institution_ids":["https://openalex.org/I201448701"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5029531416","display_name":"Navin Goyal","orcid":"https://orcid.org/0000-0002-8521-0108"},"institutions":[{"id":"https://openalex.org/I4210124949","display_name":"Microsoft Research (India)","ror":"https://ror.org/02w7f3w92","country_code":"IN","type":"company","lineage":["https://openalex.org/I1290206253","https://openalex.org/I4210124949"]}],"countries":["IN"],"is_corresponding":false,"raw_author_name":"Navin Goyal","raw_affiliation_strings":["Microsoft Research, India"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"Microsoft Research, India","institution_ids":["https://openalex.org/I4210124949"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5062910836","display_name":"Yulia Tsvetkov","orcid":"https://orcid.org/0000-0002-4634-7128"},"institutions":[{"id":"https://openalex.org/I201448701","display_name":"University of Washington","ror":"https://ror.org/00cvxb145","country_code":"US","type":"education","lineage":["https://openalex.org/I201448701"]}],"countries":["US"],"is_corresponding":false,"raw_author_name":"Yulia Tsvetkov","raw_affiliation_strings":["University of Washington, USA"],"raw_orcid":null,"affiliations":[{"raw_affiliation_string":"University of Washington, USA","institution_ids":["https://openalex.org/I201448701"]}]}],"institutions":[],"countries_distinct_count":4,"institutions_distinct_count":6,"corresponding_author_ids":[],"corresponding_institution_ids":[],"apc_list":null,"apc_paid":null,"fwci":9.772,"has_fulltext":true,"cited_by_count":7,"citation_normalized_percentile":{"value":0.97611511,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":97,"max":99},"biblio":{"volume":"13","issue":null,"first_page":"121","last_page":"141"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10181","display_name":"Natural Language Processing Techniques","score":0.9993000030517578,"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/T10181","display_name":"Natural Language Processing Techniques","score":0.9993000030517578,"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/T12090","display_name":"Language and cultural evolution","score":0.9944999814033508,"subfield":{"id":"https://openalex.org/subfields/3316","display_name":"Cultural Studies"},"field":{"id":"https://openalex.org/fields/33","display_name":"Social Sciences"},"domain":{"id":"https://openalex.org/domains/2","display_name":"Social Sciences"}},{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":0.9939000010490417,"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/computer-science","display_name":"Computer science","score":0.7818394899368286},{"id":"https://openalex.org/keywords/transformer","display_name":"Transformer","score":0.6454704403877258},{"id":"https://openalex.org/keywords/generalization","display_name":"Generalization","score":0.6220555901527405},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.5619344711303711},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.5287777185440063},{"id":"https://openalex.org/keywords/syntax","display_name":"Syntax","score":0.5157699584960938},{"id":"https://openalex.org/keywords/machine-learning","display_name":"Machine learning","score":0.37074053287506104},{"id":"https://openalex.org/keywords/mathematics","display_name":"Mathematics","score":0.08292010426521301}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7818394899368286},{"id":"https://openalex.org/C66322947","wikidata":"https://www.wikidata.org/wiki/Q11658","display_name":"Transformer","level":3,"score":0.6454704403877258},{"id":"https://openalex.org/C177148314","wikidata":"https://www.wikidata.org/wiki/Q170084","display_name":"Generalization","level":2,"score":0.6220555901527405},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.5619344711303711},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.5287777185440063},{"id":"https://openalex.org/C60048249","wikidata":"https://www.wikidata.org/wiki/Q37437","display_name":"Syntax","level":2,"score":0.5157699584960938},{"id":"https://openalex.org/C119857082","wikidata":"https://www.wikidata.org/wiki/Q2539","display_name":"Machine learning","level":1,"score":0.37074053287506104},{"id":"https://openalex.org/C33923547","wikidata":"https://www.wikidata.org/wiki/Q395","display_name":"Mathematics","level":0,"score":0.08292010426521301},{"id":"https://openalex.org/C134306372","wikidata":"https://www.wikidata.org/wiki/Q7754","display_name":"Mathematical analysis","level":1,"score":0.0},{"id":"https://openalex.org/C62520636","wikidata":"https://www.wikidata.org/wiki/Q944","display_name":"Quantum mechanics","level":1,"score":0.0},{"id":"https://openalex.org/C121332964","wikidata":"https://www.wikidata.org/wiki/Q413","display_name":"Physics","level":0,"score":0.0},{"id":"https://openalex.org/C165801399","wikidata":"https://www.wikidata.org/wiki/Q25428","display_name":"Voltage","level":2,"score":0.0}],"mesh":[],"locations_count":2,"locations":[{"id":"doi:10.1162/tacl_a_00733","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00733","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00733/2501791/tacl_a_00733.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},{"id":"pmh:oai:infoscience.epfl.ch:20.500.14299/247115","is_oa":true,"landing_page_url":"https://infoscience.epfl.ch/handle/20.500.14299/247115","pdf_url":null,"source":{"id":"https://openalex.org/S4306400487","display_name":"Infoscience (Ecole Polytechnique F\u00e9d\u00e9rale de Lausanne)","issn_l":null,"issn":null,"is_oa":true,"is_in_doaj":false,"is_core":false,"host_organization":null,"host_organization_name":null,"host_organization_lineage":[],"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":"research article"}],"best_oa_location":{"id":"doi:10.1162/tacl_a_00733","is_oa":true,"landing_page_url":"https://doi.org/10.1162/tacl_a_00733","pdf_url":"https://direct.mit.edu/tacl/article-pdf/doi/10.1162/tacl_a_00733/2501791/tacl_a_00733.pdf","source":{"id":"https://openalex.org/S2729999759","display_name":"Transactions of the Association for Computational Linguistics","issn_l":"2307-387X","issn":["2307-387X"],"is_oa":true,"is_in_doaj":true,"is_core":true,"host_organization":"https://openalex.org/P4310320244","host_organization_name":"Association for Computational Linguistics","host_organization_lineage":["https://openalex.org/P4310320244"],"host_organization_lineage_names":["Association for Computational Linguistics"],"type":"journal"},"license":"cc-by","license_id":"https://openalex.org/licenses/cc-by","version":"publishedVersion","is_accepted":true,"is_published":true,"raw_source_name":"Transactions of the Association for Computational Linguistics","raw_type":"journal-article"},"sustainable_development_goals":[],"awards":[{"id":"https://openalex.org/G6671297155","display_name":null,"funder_award_id":"CAREER","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"},{"id":"https://openalex.org/F4320308943","display_name":"Microsoft Research","ror":"https://ror.org/00d0nc645"}],"has_content":{"grobid_xml":true,"pdf":true},"content_urls":{"pdf":"https://content.openalex.org/works/W4407355147.pdf","grobid_xml":"https://content.openalex.org/works/W4407355147.grobid-xml"},"referenced_works_count":34,"referenced_works":["https://openalex.org/W1575798196","https://openalex.org/W2047706513","https://openalex.org/W2076749875","https://openalex.org/W2130942839","https://openalex.org/W2154346509","https://openalex.org/W2612690371","https://openalex.org/W2788924045","https://openalex.org/W2888329843","https://openalex.org/W2945260553","https://openalex.org/W2946794439","https://openalex.org/W2970454961","https://openalex.org/W2973154008","https://openalex.org/W2981584912","https://openalex.org/W2990138404","https://openalex.org/W3014415613","https://openalex.org/W3034503989","https://openalex.org/W3202908475","https://openalex.org/W4213350211","https://openalex.org/W4226412614","https://openalex.org/W4285174271","https://openalex.org/W4385571966","https://openalex.org/W4401042367","https://openalex.org/W6631190155","https://openalex.org/W6679436768","https://openalex.org/W6721454782","https://openalex.org/W6739901393","https://openalex.org/W6745499552","https://openalex.org/W6751575553","https://openalex.org/W6755207826","https://openalex.org/W6757635932","https://openalex.org/W6778883912","https://openalex.org/W6802147943","https://openalex.org/W6851109199","https://openalex.org/W6880003279"],"related_works":["https://openalex.org/W3162204513","https://openalex.org/W2371138613","https://openalex.org/W2048963458","https://openalex.org/W43109613","https://openalex.org/W2359952343","https://openalex.org/W2239445980","https://openalex.org/W4388713123","https://openalex.org/W2080152487","https://openalex.org/W3083152911","https://openalex.org/W3022347918"],"abstract_inverted_index":{"Abstract":[0],"Transformers":[1],"trained":[2,48],"on":[3,49,136,166],"natural":[4],"language":[5,77,183],"data":[6],"have":[7],"been":[8],"shown":[9],"to":[10,71,73,82,120,154],"exhibit":[11],"hierarchical":[12,41,74,125,151,170],"generalization":[13,91,171,181],"without":[14],"explicitly":[15],"encoding":[16],"any":[17],"structural":[18],"bias.":[19],"In":[20],"this":[21],"work,":[22],"we":[23,115],"investigate":[24],"sources":[25],"of":[26,106,144,169],"inductive":[27],"bias":[28],"in":[29,172,182],"transformer":[30],"models":[31],"and":[32,58,139,174],"their":[33],"training":[34,56,96],"that":[35,101,145],"could":[36],"cause":[37],"such":[38,63],"preference":[39,123],"for":[40,124,179],"generalization.":[42,159],"We":[43,86,127],"extensively":[44],"experiment":[45],"with":[46],"transformers":[47,83,133,173],"five":[50],"synthetic,":[51],"controlled":[52],"datasets":[53],"using":[54],"several":[55],"objectives":[57,62],"show":[59],"that,":[60],"while":[61],"as":[64],"sequence-to-sequence":[65],"modeling,":[66],"classification,":[67],"etc.,":[68],"often":[69],"fail":[70],"lead":[72],"generalization,":[75],"the":[76,95,103,109,141,167],"modeling":[78],"objective":[79],"consistently":[80],"leads":[81],"generalizing":[84],"hierarchically.":[85],"then":[87],"study":[88],"how":[89],"different":[90,112],"behaviors":[92],"emerge":[93],"during":[94],"by":[97,149],"conducting":[98],"pruning":[99],"experiments":[100],"reveal":[102],"joint":[104],"existence":[105],"subnetworks":[107],"within":[108],"model":[110],"implementing":[111],"generalizations.":[113],"Finally,":[114],"take":[116],"a":[117,129,137,150,176],"Bayesian":[118],"perspective":[119],"understand":[121],"transformers\u2019":[122],"generalization:":[126],"establish":[128],"correlation":[130],"between":[131],"whether":[132],"generalize":[134],"hierarchically":[135],"dataset":[138,146],"if":[140],"simplest":[142],"explanation":[143],"is":[147],"provided":[148],"grammar":[152],"compared":[153],"regular":[155],"grammars":[156],"exhibiting":[157],"linear":[158],"Overall,":[160],"our":[161],"work":[162],"presents":[163],"new":[164],"insights":[165],"origins":[168],"provides":[175],"theoretical":[177],"framework":[178],"studying":[180],"models.":[184]},"counts_by_year":[{"year":2026,"cited_by_count":3},{"year":2025,"cited_by_count":4}],"updated_date":"2026-05-21T06:26:12.895304","created_date":"2025-10-10T00:00:00"}