Open research, datasets, and tools for working with regulations, guidance, and compliance text.

RegNLP is a research-led initiative focused on Regulatory Natural Language Processing.
We build and share open-source resources that make regulatory and compliance documents easier to:

• Search – information retrieval and ranking for long, multi-passage regulations
• Ask – question answering and Retrieval-Augmented Generation (RAG)
• Analyse – obligation extraction, classification, and domain-specific evaluation metrics

Our aim is to support both researchers and practitioners who work with complex regulatory frameworks.

• Regulatory QA and RAG
  Multi-passage question answering and grounded answer generation over regulations and guidance.

• Regulatory information retrieval
  Dense and hybrid retrieval, learning-to-rank, and graph-aware methods for long, structured documents.

• Obligations and requirements
  Identification and classification of duties, permissions, and prohibitions in regulatory text.

• Evaluation for compliance use cases
  Metrics that go beyond generic NLP scores to capture entailment, contradiction, obligation coverage, linguistic quality, and answer stability.

A selection of core repositories under the RegNLP organisation:

• ObliQADataset
  Obligation-centric regulatory QA benchmark and data creation pipeline.

• ObliQA-MultiPassage
  Multi-passage extension of ObliQA for regulatory IR and answer generation.

• RePASs
  Regulatory Passage Answer Stability Score – a composite metric for evaluating regulatory QA answers.

• MultiPassage-RegulatoryRAG
  End-to-end RAG pipeline combining BM25, dense retrieval, RRF, and learning-to-rank.

• ObligationClassifier
  Fine-tuning LegalBERT for obligation vs. non-obligation classification in regulatory texts.

For a broader curated overview of publications, datasets, models, and events related to regulatory NLP, see the Resources section on the RegNLP website.

• Website: https://regnlp.github.io
• The site hosts information about RegNLP workshops, including calls for papers, programmes, and links to proceedings, as well as a curated resources page for the wider community.

• Use and cite the datasets, models, and code in your own research.
• Open Issues and Pull Requests to report bugs, improve documentation, or contribute new features.
• Suggest new regulatory NLP resources (datasets, benchmarks, tools) that should be highlighted.

We welcome collaboration from both academia and industry working on compliance, regtech, and legal NLP.

For general enquiries and suggestions for additional resources:

• Email: regnlp.community@gmail.com

Project-specific contact details are also listed in the individual repositories where applicable.