{"corpus_id":52967399,"paper_sha":"df2b0e26d0599ce3e70df8a9da02e51594e0e992","doi":"10.18653/v1/N19-1423","arxiv_id":"1810.04805","pmid":null,"pmcid":null,"mag_id":2963341956,"dblp_id":"conf/naacl/DevlinCLT19","acl_id":"N19-1423","title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","year":2019,"publication_date":null,"venue":"North American Chapter of the Association for Computational Linguistics","journal":{"name":null,"pages":"4171-4186","volume":null},"journal_issn":null,"journal_title":null,"publication_types":["JournalArticle","Conference"],"pubmed_pub_types":null,"s2_fields_of_study":["Computer Science"],"reference_count":63,"citation_count":111193,"influential_citation_count":21997,"is_open_access":false,"arxiv_categories":["cs.CL"],"arxiv_license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","arxiv_journal_ref":null,"mesh_headings":null,"chemicals":null,"comments_corrections":null,"source_flags":1,"s2_open_access_pdf_url":null,"s2_open_access_landing_url":null,"s2_open_access_license":null,"s2_open_access_status":null,"pmc_open_access_pdf_url":null,"pmc_open_access_landing_url":null,"pmc_open_access_license":null,"pmc_open_access_status":null,"unpaywall_open_access_pdf_url":null,"unpaywall_open_access_landing_url":null,"unpaywall_open_access_license":null,"unpaywall_open_access_status":null,"abstract":"We introduce a new language representation model called BERT, which stands for Bidirectional Encoder Representations from Transformers. Unlike recent language representation models (Peters et al., 2018a; Radford et al., 2018), BERT is designed to pre-train deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers. As a result, the pre-trained BERT model can be fine-tuned with just one additional output layer to create state-of-the-art models for a wide range of tasks, such as question answering and language inference, without substantial task-specific architecture modifications. BERT is conceptually simple and empirically powerful. It obtains new state-of-the-art results on eleven natural language processing tasks, including pushing the GLUE score to 80.5 (7.7 point absolute improvement), MultiNLI accuracy to 86.7% (4.6% absolute improvement), SQuAD v1.1 question answering Test F1 to 93.2 (1.5 point absolute improvement) and SQuAD v2.0 Test F1 to 83.1 (5.1 point absolute improvement).","claims":[{"public_id":"cl_43e054a1ab2cd062df15fdf293231f04","status":"active","text":"BERT achieves new state-of-the-art results on eleven natural language processing tasks, including GLUE, MultiNLI, SQuAD v1.1, and SQuAD v2.0.","confidence":0.98,"contributors":[],"url":"https://sah.borca.ai/claims/cl_43e054a1ab2cd062df15fdf293231f04"},{"public_id":"cl_c79702585b2fbf9fb9fdee316f55d144","status":"active","text":"BERT pre-trains deep bidirectional representations from unlabeled text by jointly conditioning on both left and right context in all layers.","confidence":0.99,"contributors":[],"url":"https://sah.borca.ai/claims/cl_c79702585b2fbf9fb9fdee316f55d144"},{"public_id":"cl_fe39370174d3f883e229dcc15d9dc952","status":"active","text":"Fine-tuning BERT with only one additional output layer yields state-of-the-art performance on a wide range of language understanding tasks without substantial task-specific architecture modifications.","confidence":0.97,"contributors":[],"url":"https://sah.borca.ai/claims/cl_fe39370174d3f883e229dcc15d9dc952"}],"concepts":[{"public_id":"co_0b4625930dbcf2951043ff76398147b8","status":"active","name":"natural language processing tasks","description":"Benchmark tasks in NLP used to evaluate model performance.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_0b4625930dbcf2951043ff76398147b8"},{"public_id":"co_12779413b1cc47e9648d75c2f0ee04fd","status":"active","name":"multinli","description":"A natural language inference benchmark with diverse premise-hypothesis pairs.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_12779413b1cc47e9648d75c2f0ee04fd"},{"public_id":"co_1323fc9ae6396dabe3ebe5108a36d950","status":"active","name":"squad v2.0","description":"A question answering benchmark that includes both answerable and unanswerable questions.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_1323fc9ae6396dabe3ebe5108a36d950"},{"public_id":"co_21051b02dca945cbb42307dc1cdcb538","status":"active","name":"bidirectional representations","description":"Text representations that incorporate information from both left and right context.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_21051b02dca945cbb42307dc1cdcb538"},{"public_id":"co_2ce7088662da1a83b9fd2bbe88364f97","status":"active","name":"fine-tuning","description":"Adapting a pretrained model to a downstream task using task-specific training.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_2ce7088662da1a83b9fd2bbe88364f97"},{"public_id":"co_9f1cc53f5c99b200ba0734e6b4aa9818","status":"active","name":"bert","description":"A transformer-based language representation model pretrained on unlabeled text and then fine-tuned for downstream NLP tasks.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_9f1cc53f5c99b200ba0734e6b4aa9818"},{"public_id":"co_a6fc79184555c7cdb92c5c4c99e79935","status":"active","name":"output layer","description":"The additional task-specific layer added on top of the pretrained model for downstream prediction.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_a6fc79184555c7cdb92c5c4c99e79935"},{"public_id":"co_a7813c9e5e73f9da9600054418f198e4","status":"active","name":"language understanding tasks","description":"Downstream natural language tasks that require semantic and contextual comprehension.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_a7813c9e5e73f9da9600054418f198e4"},{"public_id":"co_c3da7e67be83c1fe94307117381216ab","status":"active","name":"left and right context","description":"The tokens appearing before and after a target position in a sequence.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_c3da7e67be83c1fe94307117381216ab"},{"public_id":"co_c7176e29c6df274198ce93f3a6ae2d4a","status":"active","name":"glue","description":"A benchmark suite for evaluating general language understanding systems.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_c7176e29c6df274198ce93f3a6ae2d4a"},{"public_id":"co_cb5b7dad6d82dc1af553e48b718e3319","status":"active","name":"squad v1.1","description":"A question answering benchmark based on answerable questions over Wikipedia passages.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_cb5b7dad6d82dc1af553e48b718e3319"},{"public_id":"co_df54ce83357db3fa1e23c68b1a177f84","status":"active","name":"unlabeled text","description":"Text data without task-specific annotations used for pre-training.","types":[],"aliases":[],"contributors":[],"url":"https://sah.borca.ai/concepts/co_df54ce83357db3fa1e23c68b1a177f84"}],"external_ids":{"DOI":"10.18653/v1/N19-1423","ArXiv":"1810.04805","PubMed":null,"PubMedCentral":null,"MAG":2963341956,"DBLP":"conf/naacl/DevlinCLT19","ACL":"N19-1423"},"open_access":{"is_open_access":true,"pdf_url":"https://arxiv.org/pdf/1810.04805","landing_url":"https://arxiv.org/abs/1810.04805","source":"arxiv","pdf_url_source":"derived_arxiv","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","reason":null},"reference_availability":{"status":"available","references_indexed":true,"full_text_available":true,"full_text_source":"arxiv","count_basis":"semantic_scholar_metadata","extraction_status":"not_applicable","reason":null},"source":{"provider":"episteme2","base_corpus":"semantic_scholar_dump","freshness_mode":"unknown","basis":["semantic_scholar_metadata","postgres_metadata"],"limits":["paper metadata is based on indexed upstream scholarly datasets","claims and concepts are available only for extracted papers","absence of claims or concepts means no extracted graph data is available in this response"],"status":"available","degraded":false,"degraded_reasons":[],"diagnostics":{"status":"available","degraded":false,"degraded_reasons":[],"metadata_status":"available","graph_status":"available","abstract_status":"available"},"source_flags":1},"paper_id":150951,"paper_uid":"336b5649-fcbb-4a54-9bb3-319eb570353b","canonical_identity":{"paper_id":150951,"paper_uid":"336b5649-fcbb-4a54-9bb3-319eb570353b","identity_status":"available","lookup_basis":"semantic_scholar_external_id","compatibility_path":"corpus_id"},"url":"https://sah.borca.ai/papers/52967399"}