IL-PCSR (Indian Legal — Precedent & Statute Retrieval)

IL-PCSR: A dataset of Indian legal case queries annotated with relevant statutes and precedents, plus candidate pools of statutory provisions and precedent judgments for retrieval research.

---

Dataset Summary

IL-PCSR contains three configs:

• queries — case queries (judgments / case descriptions) annotated with relevant statutes and precedents.
• statutes — statutory provisions (candidate documents).
• precedents — prior judgments (candidate documents).

---

Languages

• English (Indian legal terminology and occasional code-switching).

---

License & Access

• License: CC-BY-NC-SA 4.0 (non-commercial, share-alike).
• This dataset is gated. To request access, users must complete the provided access form (the gate fields are defined in the front-matter). By requesting the dataset the user must explicitly agree to:
  • Use the dataset only for non-commercial research purposes, and
  • Not further redistribute or upload the dataset anywhere else.
• DISCLAIMER (must be accepted): the dataset is released for research purposes only; authors disclaim responsibility for any loss or damage from usage.

---

Dataset structure (detailed)

The YAML in the front-matter contains exact dataset_info and configs. Below is a readable description of the same schema.

Config: queries (default)

Splits

• train_queries: 5,017 examples
• dev_queries: 627 examples
• test_queries: 627 examples

Fields / schema

• id: string → unique id for the query.
• case_title: string → short title of the case.
• date: string → judgment date in ISO:YYYY-MM-DD format.
• jurisdiction: string → court or jurisdiction name.
• text: list[string] — tokenized/segmented text lines or paragraphs (stored as a list to preserve segments).
• rhetorical_roles: list[string] → role labels for each segment (aligned with text segments).
  • [Facts, Issue, Argument by Petitioner, Argument by Respondent, Statute Analysis, Precedent Analysis, Court Disclosure, Court Reasoning, NONE]
• relevant_statutes: list[string] → human-readable statute names referenced as relevant.
• relevant_statute_ids: list[string] → canonical IDs matching the statutes config id field.
• relevant_precedents: list[string] → human-readable precedent case titles.
• relevant_precedent_ids: list[string] → canonical IDs matching the precedents config id field.

Notes

• relevant_statute_ids and relevant_precedent_ids are the fields used to build ground-truth relevance sets (qrels) for retrieval evaluation.

---

Config: statutes

Splits

• statute_candidates: 936 examples

Fields

• id: string → unique provision id (matches relevant_statute_ids in queries).
• provision_name: string → statute/provision title.
• text: list[string] → provision text (segmented).

---

Config: precedents

Splits

• precedent_candidates: 3,183 examples

Fields / schema - Same as queries

Data provenance & preprocessing

• Sources: Publicly available Indian judgments and statutory texts (collected from public repositories and court websites).
• Curation steps (summary):
  • Document collection and deduplication.
  • Minimal normalization (unicode normalisation, whitespace clean-up).
  • Masking legal named entities and citations
  • Rhetorical segments (Facts, Issues, Reasoning, Judgment) provided according to IndianKanoon in rhetorical_roles.
  • Linking queries to statutes and precedents (producing the relevant_* fields).
• PII & legal caution: Users are responsible for additional PII removal or compliance checks; the dataset maintainers provide the data "as curated" and do not guarantee absence of sensitive information.

---

How to use

All of the examples assume you have access (i.e., gate accepted).

1) Load from Hugging Face Hub

from datasets import load_dataset

# Load queries config (default)
ds_queries = load_dataset("Exploration-Lab/IL-PCSR", name="queries")

# Access splits using the explicit split names
train_q = ds_queries["train_queries"]
dev_q   = ds_queries["dev_queries"]
test_q  = ds_queries["test_queries"]

# Example: inspect first query
print(train_q[0])

2) Load specific configs

# Statutes (candidate pool)
ds_statutes = load_dataset("Exploration-Lab/IL-PCSR", name="statutes")
statute_pool = ds_statutes["statute_candidates"]
print(statute_pool[0])

# Precedents (candidate pool)
ds_precedents = load_dataset("Exploration-Lab/IL-PCSR", name="precedents")
precedent_pool = ds_precedents["precedent_candidates"]
print(precedent_pool[0])

3) Local development (loading from a local repo folder)

# If you're testing locally (folder contains dataset script + README.md + data/)
ds_local_queries = load_dataset("</path/to/local/IL-PCSR>", name="queries")
print(ds_local_queries["train_queries"][0])

4) Building qrels (ground-truth) for retrieval evaluation

# Example: build a qrels-like dict mapping query id -> set(candidate_ids)
def build_qrels(queries_split):
    qrels = {}
    for ex in queries_split:
        qid = ex["id"]
        # combine statute ids and precedent ids if you want a single candidate set
        relevant = set(ex.get("relevant_statute_ids", []) + ex.get("relevant_precedent_ids", []))
        qrels[qid] = list(relevant)
    return qrels

qrels_train = build_qrels(train_q)
print("sample qrels:", list(qrels_train.items())[:2])

Citation

@inproceedings{il-pcsr2025,
  title = "IL-PCSR: Legal Corpus for Prior Case and Statute Retrieval",
  author = "Paul, Shounak and Ghumare, Dhananjay and Goyal, Pawan and Ghosh, Saptarshi and Modi, Ashutosh"
  booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing",
  month = nov,
  year = "2025",
  address = "Suzhou, China",
  publisher = "Association for Computational Linguistics",
  note = "To Appear"
}