LegalReasoningBench / README.md
archon159's picture
Update README.md
9ba7893 verified
|
Raw History Blame Contribute Delete
3.29 kB
metadata
license: cc-by-4.0
language:
  - en
tags:
  - legal
  - reasoning
  - benchmark
  - ecthr
configs:
  - config_name: default
    data_files:
      - split: test
        path: cases.jsonl

LegalReasoningBench

A benchmark of structured gold-standard annotations for European Court of Human Rights (ECtHR) Article 11 (freedom of assembly) cases — covering case-level conclusions, six doctrinal element conclusions, and fact-to-element mappings (List 1 / List 2 / List 3).

Accepted at the NeurIPS 2026 Datasets and Benchmarks Track.

Code and reproduction scripts: github.com/archon159/LegalReasoningBench

Quick start

from datasets import load_dataset

ds = load_dataset("archon159/LegalReasoningBench", split="test")
print(ds[0])

The dataset has 375 examples, one per case.

Schema

{
  "case_name": "case_274",
  "atomic_facts": [
    {"fact_id": "F001", "text": "...", "source_paragraph": 3, "section_path": ["THE FACTS"]},
    ...
  ],
  "conclusion": {
    "violated": true,
    "elements": {
      "scope":             {"conclusion": "yes",  "discussed": "explicit"},
      "interference":      {"conclusion": "yes",  "discussed": "explicit"},
      "prescribed_by_law": {"conclusion": "yes",  "discussed": "implicit"},
      "legitimate_aim":    {"conclusion": "yes",  "discussed": "explicit",
                            "aims": ["prevention_of_disorder", ...]},
      "necessity":         {"conclusion": "no",   "discussed": "explicit"},
      "positive_obligation": {"conclusion": null, "discussed": "not_discussed"}
    }
  },
  "fact_mapping": {
    "list1": {"scope": ["F003", "F012", ...], "interference": [...], ...},
    "list2": {"scope": [...], ...},
    "list3": ["F005", "F012", ...]
  }
}
  • atomic_facts — structured facts extracted from the judgment (model input).
  • conclusion.violated — case-level Article 11 violation finding.
  • conclusion.elements[el].conclusion — "yes" / "no" / null (not discussed).
  • conclusion.elements[el].discussed — "explicit" / "implied" / "implicit" / "not_discussed".
  • fact_mapping.listN — fact IDs referencing atomic_facts[*].fact_id. Subset constraint: List 3 ⊆ List 2 ⊆ List 1.

Files

  • cases.jsonl — primary file, one case per line.
  • raw/ — supporting per-case JSONs (also used to regenerate cases.jsonl):
    • compiled_gold_standard/ — gold annotations (one file per case).
    • atomic_facts/ — atomic facts (one file per case).
    • included_all_case_names.txt — list of case IDs.

License

The annotations are released under CC BY 4.0. The underlying ECtHR judgment texts are public-domain documents from HUDOC.

Citation

A full citation will be added once the NeurIPS 2026 proceedings are available. Until then, please cite as:

@inproceedings{paul2026legalreasoningbench,
  title     = {LegalReasoningBench: A Benchmark of Structured Annotations for ECtHR Article 11 Cases},
  author    = {Paul, Shounak and Lee, Seungeon and Engel, Christoph and Gummadi, Krishna P.},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS), Datasets and Benchmarks Track},
  year      = {2026}
}