{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/when-does-pretraining-help-assessing-self","title":"When Does Pretraining Help? Assessing Self-Supervised Learning for Law and the CaseHOLD Dataset","arxiv_id":"2104.08671","date":"2021-04-18","proceeding":null,"authors":["Lucia Zheng","Neel Guha","Brandon R. Anderson","Peter Henderson","Daniel E. Ho"],"abstract":"While self-supervised learning has made rapid advances in natural language processing, it remains unclear when researchers should engage in resource-intensive domain-specific pretraining (domain pretraining). The law, puzzlingly, has yielded few documented instances of substantial gains to domain pretraining in spite of the fact that legal language is widely seen to be unique. We hypothesize that these existing results stem from the fact that existing legal NLP tasks are too easy and fail to meet conditions for when domain pretraining can help. To address this, we first present CaseHOLD (Case Holdings On Legal Decisions), a new dataset comprised of over 53,000+ multiple choice questions to identify the relevant holding of a cited case. This dataset presents a fundamental task to lawyers and is both legally meaningful and difficult from an NLP perspective (F1 of 0.4 with a BiLSTM baseline). Second, we assess performance gains on CaseHOLD and existing legal NLP datasets. While a Transformer architecture (BERT) pretrained on a general corpus (Google Books and Wikipedia) improves performance, domain pretraining (using corpus of approximately 3.5M decisions across all courts in the U.S. that is larger than BERT's) with a custom legal vocabulary exhibits the most substantial performance gains with CaseHOLD (gain of 7.2% on F1, representing a 12% improvement on BERT) and consistent performance gains across two other legal tasks. Third, we show that domain pretraining may be warranted when the task exhibits sufficient similarity to the pretraining corpus: the level of performance increase in three legal tasks was directly tied to the domain specificity of the task. Our findings inform when researchers should engage resource-intensive pretraining and show that Transformer-based architectures, too, learn embeddings suggestive of distinct legal language.","url_abs":"https://arxiv.org/abs/2104.08671v3","url_pdf":"https://arxiv.org/pdf/2104.08671v3.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"when-does-pretraining-help-assessing-self","repo_url":"https://github.com/reglab/casehold","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}},{"paper_slug":"when-does-pretraining-help-assessing-self","repo_url":"https://github.com/trusthlt/privacy-legal-nlp-lm","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"jax","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"multiple-choice","task_name":"Multiple-choice"},{"task_slug":"question-answering","task_name":"Question Answering"},{"task_slug":"self-supervised-learning","task_name":"Self-Supervised Learning"},{"task_slug":"specificity","task_name":"Specificity"},{"task_slug":"text-classification","task_name":"Text Classification"}],"methods":[{"method_slug":"absolute-position-encodings","method_name":"Absolute Position Encodings"},{"method_slug":"adam","method_name":"Adam"},{"method_slug":"attention","method_name":"Attention"},{"method_slug":"bpe","method_name":"BPE"},{"method_slug":"bilstm","method_name":"BiLSTM"},{"method_slug":"dense-connections","method_name":"Dense Connections"},{"method_slug":"dropout","method_name":"Dropout"},{"method_slug":"lstm","method_name":"LSTM"},{"method_slug":"label-smoothing","method_name":"Label Smoothing"},{"method_slug":"layer-normalization","method_name":"Layer Normalization"},{"method_slug":"linear-layer","method_name":"Linear Layer"},{"method_slug":"multi-head-attention","method_name":"Multi-Head Attention"},{"method_slug":"position-wise-feed-forward-layer","method_name":"Position-Wise Feed-Forward Layer"},{"method_slug":"residual-connection","method_name":"Residual Connection"},{"method_slug":"sigmoid-activation","method_name":"Sigmoid Activation"},{"method_slug":"softmax","method_name":"Softmax"},{"method_slug":"tanh-activation","method_name":"Tanh Activation"},{"method_slug":"transformer","method_name":"Transformer"}],"datasets_introduced":[{"slug":"casehold","name":"CaseHOLD","full_name":"Case Holdings On Legal Decisions"},{"slug":"overruling","name":"Overruling","full_name":""}],"methods_introduced":[],"results":[{"leaderboard":"/sota/question-answering-on-casehold","task":"Question Answering","dataset":"CaseHOLD","model":"Custom Legal-BERT","rank_in_archive_order":1,"of":3,"metrics":{"Macro F1 (10-fold)":"69.5"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-casehold","task":"Question Answering","dataset":"CaseHOLD","model":"Legal-BERT","rank_in_archive_order":2,"of":3,"metrics":{"Macro F1 (10-fold)":"68.0"},"uses_additional_data":false},{"leaderboard":"/sota/question-answering-on-casehold","task":"Question Answering","dataset":"CaseHOLD","model":"BERT","rank_in_archive_order":3,"of":3,"metrics":{"Macro F1 (10-fold)":"61.3"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-overruling","task":"Text Classification","dataset":"Overruling","model":"Custom Legal-BERT","rank_in_archive_order":1,"of":3,"metrics":{"F1(10-fold)":"97.4"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-overruling","task":"Text Classification","dataset":"Overruling","model":"Legal-BERT","rank_in_archive_order":2,"of":3,"metrics":{"F1(10-fold)":"96.3"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-overruling","task":"Text Classification","dataset":"Overruling","model":"BERT","rank_in_archive_order":3,"of":3,"metrics":{"F1(10-fold)":"95.8"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-terms-of-service","task":"Text Classification","dataset":"Terms of Service","model":"Custom Legal-BERT","rank_in_archive_order":1,"of":3,"metrics":{"F1(10-fold)":"78.7"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-terms-of-service","task":"Text Classification","dataset":"Terms of Service","model":"Legal-BERT","rank_in_archive_order":2,"of":3,"metrics":{"F1(10-fold)":"75.0"},"uses_additional_data":false},{"leaderboard":"/sota/text-classification-on-terms-of-service","task":"Text Classification","dataset":"Terms of Service","model":"BERT","rank_in_archive_order":3,"of":3,"metrics":{"F1(10-fold)":"72.2"},"uses_additional_data":false}],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2104.08671","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2104.08671"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/reglab/casehold","reach":{"status":"ok","spdx":"Apache-2.0"}},{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/trusthlt/privacy-legal-nlp-lm","reach":{"status":"ok","spdx":"Apache-2.0"}}],"summary":{"unverified":9},"by_repo_kind":{"official":{"samples":1,"ran":0,"repositories":1},"listed":{"samples":8,"ran":0,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"e1f72af0ac263f37","entry":"apply_threshs","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"downstream_tasks/utils.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/downstream_tasks/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"e1f72af0ac263f37"}},{"code_sha256_prefix":"f252d20c754d3bcf","entry":"build_dataset","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"downstream_tasks/data_loader.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/downstream_tasks/data_loader.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"f252d20c754d3bcf"}},{"code_sha256_prefix":"68377121bfc285a9","entry":"convert_examples_to_features","repo":"reglab/casehold","repo_kind":"official","path":"multiple_choice/utils_multiple_choice.py","file_url":"https://github.com/reglab/casehold/blob/HEAD/multiple_choice/utils_multiple_choice.py","link_basis":"first_harvest_node","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"68377121bfc285a9"}},{"code_sha256_prefix":"36a781e6d5afa906","entry":"create_instance_from_document","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"data_prepare/utils.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/data_prepare/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"36a781e6d5afa906"}},{"code_sha256_prefix":"45127374f48c9723","entry":"equal","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"pretrain/utils.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/pretrain/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"45127374f48c9723"}},{"code_sha256_prefix":"efbd156ad839d2bf","entry":"generate_random_next","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"data_prepare/utils.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/data_prepare/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"efbd156ad839d2bf"}},{"code_sha256_prefix":"851b9497b3ad4904","entry":"multihot","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"downstream_tasks/data_loader.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/downstream_tasks/data_loader.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"851b9497b3ad4904"}},{"code_sha256_prefix":"14ea19820452ef31","entry":"split_corpus","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"downstream_tasks/data_loader.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/downstream_tasks/data_loader.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"14ea19820452ef31"}},{"code_sha256_prefix":"a58a2a6534b81101","entry":"tune_threshs","repo":"trusthlt/privacy-legal-nlp-lm","repo_kind":"listed","path":"downstream_tasks/utils.py","file_url":"https://github.com/trusthlt/privacy-legal-nlp-lm/blob/HEAD/downstream_tasks/utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"Apache-2.0","inline_ok":true,"mcp_get_code":{"code_sha256":"a58a2a6534b81101"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}