README.md

---
license: cc-by-4.0
task_categories:
- text-classification
language:
- en
tags:
- Text Classification
- Named Entity Recognition
- Structured Extraction
- Legal NLP
- International Law
---

IntLawNER is an English-language named entity recognition dataset for legal
instrument and citation references in international-law documents. It contains
2,987 gold sentences drawn from three public legal-document collections.

## Sources

| Code | Collection | Gold sentences |
| --- | --- | ---: |
| `CD-ICJ` | International Court of Justice | 1,045 |
| `ECtHR` | European Court of Human Rights | 1,149 |
| `UNSC` | UN Security Council | 793 |

The raw source documents are not bundled. Collection URLs and acquisition
metadata are recorded in `data/corpus_manifest.json`. Document provenance is
retained in each record's `corpus` and `doc_id` fields.


## Annotation Schema

Each line is a JSON object. Core fields are:

```json
{
  "sent_id": "stable sentence identifier",
  "text": "sentence text",
  "corpus": "CD-ICJ, ECtHR, or UNSC",
  "doc_id": "source document identifier",
  "label": [
    {
      "start": 0,
      "end": 10,
      "text": "exact span",
      "labels": ["IL_PROVISION"]
    }
  ]
}
```

## Entity Distribution

| Label | Count |
| --- | ---: |
| `IL_PROVISION` | 3,026 |
| `IL_TREATY` | 1,641 |
| `IL_RESOLUTION` | 1,322 |
| `IL_DOMESTIC_LAW` | 1,208 |
| `IL_CASE_CITATION` | 622 |
| `IL_SOFT_LAW` | 192 |
| `IL_COURT_INSTR` | 83 |

Definitions and concise examples are provided in the main README and the
evaluation prompt glossary under `src/evaluation/`.

## Intended Uses

- Training and evaluating legal-domain NER systems.
- Studying references among international, domestic, judicial, and soft-law instruments.
- Comparing rule-based, transformer, and prompted-LLM extraction methods.

The dataset is not legal advice and should not be used as the sole basis for a
legal decision. Performance on these three collections does not establish
performance on other jurisdictions, languages, document types, or time periods.

## Licensing

The original annotation layer is CC BY 4.0. Source documents retain their
original terms.