Pile of Law: Learning Responsible Data Filtering from the Law and a 256GB Open-Source Legal Dataset
Advances in Neural Information Processing Systems 35, Datasets and Benchmarks Track
What it adds
Pairs a legal and administrative training corpus with research on context-sensitive privacy and toxicity filtering.
Read with this limitation
English and US material dominate. Source licenses vary and sensitive or copyrighted material may remain. Filtering experiments are exploratory; the original pretraining software is not fully released. Public access does not imply unrestricted reuse.
- Jurisdiction tags
- United States · Canada · European Union
- Language
- English
Research artifacts
Resources linked by the authors.
Follow the source
Selected statements located in the linked primary sources. These results have not been independently reproduced here.
The paper describes an approximately 256GB English-language legal and administrative corpus assembled from 35 sources, including judgments, filings, legislation, contracts and agency documents.
2 Pile of LawThe experiments examine learning contextual privacy rules from legal text and comparing toxicity filters with the treatment of language in legal records.
4 What Can We Learn from Legal Text?; sections 4.1–4.2
Connected research
Related question
MultiLegalPile: A 689GB Multilingual Legal Corpus
Both assemble legal corpora for language-model research; they differ in language coverage and research emphasis.
Related question
FairLex: A Multilingual Benchmark for Evaluating Fairness in Legal Text Processing
Pile of Law examines data filtering; FairLex examines fairness evaluation. Both study how legal datasets shape model behaviour.
Model evaluated by
SG-LegalCite: A Principle-Augmented Benchmark for Legal Citation Retrieval in Singapore Law
Includes Pile-of-Law BERT as a retrieval baseline. This connects model provenance to the new evaluation; it does not reproduce the Pile of Law study.
Inspect the sourceEntry status, source edition & rights
- Index entry
- Source-linked summary, prepared with agent assistance. Independent scholarly review is pending.
- Archive admission
- This bibliographic entry does not confer admission to the versioned archive.
- Edition inspected
- 10.52202/068431-2119
- Reading coverage
- Published NeurIPS paper sections 1–4, plus supplement sections B, D and G inspected. English is the corpus focus; jurisdiction facets identify documented examples, not exhaustive coverage. Author names follow the PDF. Artifact availability checked; no independent reproduction.
- Rights
- External references only. Papers and research artifacts remain with their original hosts; their own terms apply.
- Entry updated
- 2026-10-15
Source-linked summary · independent assessment pending.