{"id":"https://openalex.org/W2953337107","doi":"https://doi.org/10.18653/v1/w19-4828","title":"What Does BERT Look at? An Analysis of BERT\u2019s Attention","display_name":"What Does BERT Look at? An Analysis of BERT\u2019s Attention","publication_year":2019,"publication_date":"2019-01-01","ids":{"openalex":"https://openalex.org/W2953337107","doi":"https://doi.org/10.18653/v1/w19-4828","mag":"2953337107"},"language":"en","primary_location":{"id":"doi:10.18653/v1/w19-4828","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/w19-4828","pdf_url":"https://www.aclweb.org/anthology/W19-4828.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":"Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP","raw_type":"proceedings-article"},"type":"preprint","indexed_in":["arxiv","crossref","datacite"],"open_access":{"is_oa":true,"oa_status":"gold","oa_url":"https://www.aclweb.org/anthology/W19-4828.pdf","any_repository_has_fulltext":true},"authorships":[{"author_position":"first","author":{"id":"https://openalex.org/A5049064405","display_name":"Kevin B. Clark","orcid":"https://orcid.org/0000-0003-0544-2150"},"institutions":[{"id":"https://openalex.org/I2252078561","display_name":"Meta (Israel)","ror":"https://ror.org/02388em19","country_code":"IL","type":"company","lineage":["https://openalex.org/I2252078561","https://openalex.org/I4210114444"]},{"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":["IL","US"],"is_corresponding":true,"raw_author_name":"Kevin Clark","raw_affiliation_strings":["Computer Science Department, Stanford University  Facebook AI Research","Stanford University ()"],"affiliations":[{"raw_affiliation_string":"Computer Science Department, Stanford University  Facebook AI Research","institution_ids":["https://openalex.org/I2252078561"]},{"raw_affiliation_string":"Stanford University ()","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5088072227","display_name":"Urvashi Khandelwal","orcid":null},"institutions":[{"id":"https://openalex.org/I2252078561","display_name":"Meta (Israel)","ror":"https://ror.org/02388em19","country_code":"IL","type":"company","lineage":["https://openalex.org/I2252078561","https://openalex.org/I4210114444"]},{"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":["IL","US"],"is_corresponding":false,"raw_author_name":"Urvashi Khandelwal","raw_affiliation_strings":["Computer Science Department, Stanford University  Facebook AI Research","Stanford University ()"],"affiliations":[{"raw_affiliation_string":"Computer Science Department, Stanford University  Facebook AI Research","institution_ids":["https://openalex.org/I2252078561"]},{"raw_affiliation_string":"Stanford University ()","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"middle","author":{"id":"https://openalex.org/A5024311574","display_name":"Omer Levy","orcid":"https://orcid.org/0000-0001-7300-8191"},"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"]},{"id":"https://openalex.org/I2252078561","display_name":"Meta (Israel)","ror":"https://ror.org/02388em19","country_code":"IL","type":"company","lineage":["https://openalex.org/I2252078561","https://openalex.org/I4210114444"]}],"countries":["IL","US"],"is_corresponding":false,"raw_author_name":"Omer Levy","raw_affiliation_strings":["Computer Science Department, Stanford University  Facebook AI Research","Stanford University ()"],"affiliations":[{"raw_affiliation_string":"Computer Science Department, Stanford University  Facebook AI Research","institution_ids":["https://openalex.org/I2252078561"]},{"raw_affiliation_string":"Stanford University ()","institution_ids":["https://openalex.org/I97018004"]}]},{"author_position":"last","author":{"id":"https://openalex.org/A5046006076","display_name":"Christopher D. Manning","orcid":"https://orcid.org/0000-0001-6155-649X"},"institutions":[{"id":"https://openalex.org/I2252078561","display_name":"Meta (Israel)","ror":"https://ror.org/02388em19","country_code":"IL","type":"company","lineage":["https://openalex.org/I2252078561","https://openalex.org/I4210114444"]}],"countries":["IL"],"is_corresponding":false,"raw_author_name":"Christopher D. Manning","raw_affiliation_strings":["Computer Science Department, Stanford University  Facebook AI Research","Facebook"],"affiliations":[{"raw_affiliation_string":"Computer Science Department, Stanford University  Facebook AI Research","institution_ids":["https://openalex.org/I2252078561"]},{"raw_affiliation_string":"Facebook","institution_ids":["https://openalex.org/I2252078561"]}]}],"institutions":[],"countries_distinct_count":2,"institutions_distinct_count":4,"corresponding_author_ids":["https://openalex.org/A5049064405"],"corresponding_institution_ids":["https://openalex.org/I2252078561","https://openalex.org/I97018004"],"apc_list":null,"apc_paid":null,"fwci":8.7562115,"has_fulltext":true,"cited_by_count":79,"citation_normalized_percentile":{"value":0.97909845,"is_in_top_1_percent":false,"is_in_top_10_percent":true},"cited_by_percentile_year":{"min":90,"max":100},"biblio":{"volume":null,"issue":null,"first_page":"276","last_page":"286"},"is_retracted":false,"is_paratext":false,"is_xpac":false,"primary_topic":{"id":"https://openalex.org/T10028","display_name":"Topic Modeling","score":1.0,"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":1.0,"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":0.9998999834060669,"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/T13629","display_name":"Text Readability and Simplification","score":0.9951000213623047,"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":0.7880911827087402},{"id":"https://openalex.org/keywords/sentence","display_name":"Sentence","score":0.6724190711975098},{"id":"https://openalex.org/keywords/artificial-intelligence","display_name":"Artificial intelligence","score":0.6421176195144653},{"id":"https://openalex.org/keywords/natural-language-processing","display_name":"Natural language processing","score":0.633188784122467},{"id":"https://openalex.org/keywords/coreference","display_name":"Coreference","score":0.60735684633255},{"id":"https://openalex.org/keywords/syntax","display_name":"Syntax","score":0.5755707621574402},{"id":"https://openalex.org/keywords/classifier","display_name":"Classifier (UML)","score":0.5419538617134094},{"id":"https://openalex.org/keywords/noun","display_name":"Noun","score":0.5086506009101868},{"id":"https://openalex.org/keywords/linguistics","display_name":"Linguistics","score":0.44898444414138794},{"id":"https://openalex.org/keywords/resolution","display_name":"Resolution (logic)","score":0.11952811479568481}],"concepts":[{"id":"https://openalex.org/C41008148","wikidata":"https://www.wikidata.org/wiki/Q21198","display_name":"Computer science","level":0,"score":0.7880911827087402},{"id":"https://openalex.org/C2777530160","wikidata":"https://www.wikidata.org/wiki/Q41796","display_name":"Sentence","level":2,"score":0.6724190711975098},{"id":"https://openalex.org/C154945302","wikidata":"https://www.wikidata.org/wiki/Q11660","display_name":"Artificial intelligence","level":1,"score":0.6421176195144653},{"id":"https://openalex.org/C204321447","wikidata":"https://www.wikidata.org/wiki/Q30642","display_name":"Natural language processing","level":1,"score":0.633188784122467},{"id":"https://openalex.org/C28076734","wikidata":"https://www.wikidata.org/wiki/Q63087","display_name":"Coreference","level":3,"score":0.60735684633255},{"id":"https://openalex.org/C60048249","wikidata":"https://www.wikidata.org/wiki/Q37437","display_name":"Syntax","level":2,"score":0.5755707621574402},{"id":"https://openalex.org/C95623464","wikidata":"https://www.wikidata.org/wiki/Q1096149","display_name":"Classifier (UML)","level":2,"score":0.5419538617134094},{"id":"https://openalex.org/C121934690","wikidata":"https://www.wikidata.org/wiki/Q1084","display_name":"Noun","level":2,"score":0.5086506009101868},{"id":"https://openalex.org/C41895202","wikidata":"https://www.wikidata.org/wiki/Q8162","display_name":"Linguistics","level":1,"score":0.44898444414138794},{"id":"https://openalex.org/C138268822","wikidata":"https://www.wikidata.org/wiki/Q1051925","display_name":"Resolution (logic)","level":2,"score":0.11952811479568481},{"id":"https://openalex.org/C138885662","wikidata":"https://www.wikidata.org/wiki/Q5891","display_name":"Philosophy","level":0,"score":0.0}],"mesh":[],"locations_count":4,"locations":[{"id":"doi:10.18653/v1/w19-4828","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/w19-4828","pdf_url":"https://www.aclweb.org/anthology/W19-4828.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":"Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP","raw_type":"proceedings-article"},{"id":"pmh:oai:arXiv.org:1906.04341","is_oa":true,"landing_page_url":"http://arxiv.org/abs/1906.04341","pdf_url":"https://arxiv.org/pdf/1906.04341","source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":"","raw_type":"text"},{"id":"mag:2953337107","is_oa":true,"landing_page_url":"https://arxiv.org/pdf/1906.04341.pdf","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":"arXiv (Cornell University)","raw_type":null},{"id":"doi:10.48550/arxiv.1906.04341","is_oa":true,"landing_page_url":"https://doi.org/10.48550/arxiv.1906.04341","pdf_url":null,"source":{"id":"https://openalex.org/S4306400194","display_name":"arXiv (Cornell University)","issn_l":null,"issn":null,"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":null,"is_accepted":false,"is_published":null,"raw_source_name":null,"raw_type":"article"}],"best_oa_location":{"id":"doi:10.18653/v1/w19-4828","is_oa":true,"landing_page_url":"https://doi.org/10.18653/v1/w19-4828","pdf_url":"https://www.aclweb.org/anthology/W19-4828.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":"Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP","raw_type":"proceedings-article"},"sustainable_development_goals":[{"id":"https://metadata.un.org/sdg/4","display_name":"Quality Education","score":0.8500000238418579}],"awards":[],"funders":[],"has_content":{"pdf":true,"grobid_xml":true},"content_urls":{"pdf":"https://content.openalex.org/works/W2953337107.pdf","grobid_xml":"https://content.openalex.org/works/W2953337107.grobid-xml"},"referenced_works_count":38,"referenced_works":["https://openalex.org/W1632114991","https://openalex.org/W2129657639","https://openalex.org/W2152825437","https://openalex.org/W2155069789","https://openalex.org/W2170973209","https://openalex.org/W2250539671","https://openalex.org/W2251064706","https://openalex.org/W2563574619","https://openalex.org/W2594633041","https://openalex.org/W2605717780","https://openalex.org/W2799051177","https://openalex.org/W2799124508","https://openalex.org/W2888922637","https://openalex.org/W2893141505","https://openalex.org/W2910243263","https://openalex.org/W2912126238","https://openalex.org/W2912351236","https://openalex.org/W2939556020","https://openalex.org/W2945767825","https://openalex.org/W2946417913","https://openalex.org/W2951025380","https://openalex.org/W2951528897","https://openalex.org/W2953369973","https://openalex.org/W2962739339","https://openalex.org/W2962776659","https://openalex.org/W2962784628","https://openalex.org/W2962788148","https://openalex.org/W2962811598","https://openalex.org/W2962822108","https://openalex.org/W2963341956","https://openalex.org/W2963403868","https://openalex.org/W2963651521","https://openalex.org/W2963661253","https://openalex.org/W2963751529","https://openalex.org/W2963951265","https://openalex.org/W2964222268","https://openalex.org/W2964303116","https://openalex.org/W2964308564"],"related_works":["https://openalex.org/W2963341956","https://openalex.org/W2963403868","https://openalex.org/W2972324944","https://openalex.org/W2965373594","https://openalex.org/W2963310665","https://openalex.org/W2970597249","https://openalex.org/W2946417913","https://openalex.org/W2787560479","https://openalex.org/W2964308564","https://openalex.org/W2946359678","https://openalex.org/W2996428491","https://openalex.org/W2910243263","https://openalex.org/W2250539671","https://openalex.org/W2064675550","https://openalex.org/W1522301498","https://openalex.org/W2970565456","https://openalex.org/W2951528897","https://openalex.org/W2991265431","https://openalex.org/W3007152172","https://openalex.org/W3152409010"],"abstract_inverted_index":{"Large":[0],"pre-trained":[1,65],"neural":[2],"networks":[3],"such":[4,77],"as":[5,78],"BERT":[6],"have":[7],"had":[8],"great":[9],"recent":[10,34],"success":[11],"in":[12,95,164],"NLP,":[13],"motivating":[14],"a":[15],"growing":[16],"body":[17],"of":[18,23,64,115,130,133,136],"research":[19],"investigating":[20],"what":[21],"aspects":[22],"language":[24,42],"they":[25],"are":[26],"able":[27],"to":[28,53,70,80,112,126,155],"learn":[29],"from":[30],"unlabeled":[31],"data.":[32],"Most":[33],"analysis":[35],"has":[36],"focused":[37],"on":[38],"model":[39,43],"outputs":[40],"(e.g.,":[41,49],"surprisal)":[44],"or":[45,86],"internal":[46],"vector":[47],"representations":[48],"probing":[50,150],"classifiers).":[51],"Complementary":[52],"these":[54],"works,":[55],"we":[56,121,146],"propose":[57,147],"methods":[58],"for":[59],"analyzing":[60],"the":[61,90,96,127],"attention":[62,73,108],"mechanisms":[63],"models":[66],"and":[67,117,138,152],"apply":[68],"them":[69],"BERT.":[71],"BERT's":[72,165],"heads":[74,94,109,123],"exhibit":[75],"patterns":[76],"attending":[79,88],"delimiter":[81],"tokens,":[82],"specific":[83],"positional":[84],"offsets,":[85],"broadly":[87],"over":[89],"whole":[91],"sentence,":[92],"with":[93,141],"same":[97],"layer":[98],"often":[99],"exhibiting":[100],"similar":[101],"behaviors.":[102],"We":[103],"further":[104,156],"show":[105],"that":[106,124,158],"certain":[107],"correspond":[110],"well":[111],"linguistic":[113],"notions":[114],"syntax":[116],"coreference.":[118],"For":[119],"example,":[120],"find":[122],"attend":[125],"direct":[128],"objects":[129,135],"verbs,":[131],"determiners":[132],"nouns,":[134],"prepositions,":[137],"coreferent":[139],"mentions":[140],"remarkably":[142],"high":[143],"accuracy.":[144],"Lastly,":[145],"an":[148],"attention-based":[149],"classifier":[151],"use":[153],"it":[154],"demonstrate":[157],"substantial":[159],"syntactic":[160],"information":[161],"is":[162],"captured":[163],"attention.":[166]},"counts_by_year":[{"year":2025,"cited_by_count":2},{"year":2024,"cited_by_count":1},{"year":2023,"cited_by_count":20},{"year":2022,"cited_by_count":3},{"year":2021,"cited_by_count":16},{"year":2020,"cited_by_count":26},{"year":2019,"cited_by_count":11}],"updated_date":"2026-02-09T09:26:11.010843","created_date":"2025-10-10T00:00:00"}