File size: 2,967 Bytes
e93dc3e | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 | ---
license: mit
task_categories:
- text-classification
- structure-prediction
language:
- en
tags:
- legal
- echr
- annotated-corpus
size_categories:
- n<1K
---
# ECHR Annotated Corpus
A corpus of 289 European Court of Human Rights judgments (English) with
hierarchical structural annotations produced via an AI-assisted 4-task pipeline.
## Contents
Each case includes:
| File | Description |
|------|-------------|
| `meta.json` | Case metadata: docname, date, respondent, doctype, ECLI, importance |
| `paragraphs.json` | Raw paragraphs extracted from HUDOC HTML |
| `html.html` | Original HUDOC HTML |
| `state.json` | Annotation state and per-task costs |
| `task1/` | L1 heading detection (suggestions, decisions, final) |
| `task2/` | Quote and numbered-paragraph detection |
| `task3/` | Sub-heading classification |
| `task4/` | 5-segment mapping (preamble, facts, law, conclusion, post-conclusion) |
The annotation pipeline produces a unified hierarchy:
```
segments[] → headings[] (L1-L5) → numbered paragraphs[] → quotes[]
```
Task internals are preserved for reproducibility — anyone can rebuild the final
hierarchy or re-run downstream analysis.
## Annotation methodology
1. **L1 heading detection** — regex + AI for canonical sections (INTRODUCTION,
THE FACTS, THE LAW, FOR THESE REASONS, etc.)
2. **Quote detection** — numbered-paragraph spine + between-spine classification
3. **Sub-heading classification** — Claude Haiku assigns L2-L6 levels per L1
4. **Segment mapping** — deterministic 5-segment mapping from L1 headings
Each task has explicit `suggestions` (AI/deterministic) → `decisions` (human
review) → `final` (committed) provenance.
## Loading the dataset
```bash
git clone <project-repo> echr-project
cd echr-project
pip install -r requirements.txt
# Pull the corpus from HuggingFace
python scripts/bootstrap.py
# Tell the apps where to read from
export ECHR_DATA_DIR=$(pwd)/data
# Start the viewer
cd experiments/viewer && python server.py
# Open http://127.0.0.1:5092
```
## Schema
### `meta.json`
```json
{
"itemid": "001-249367",
"docname": "CASE OF MAKKI v. DENMARK",
"judgementdate": "2025-12-15",
"respondent": "DNK",
"doctypebranch": "CHAMBER",
"importance": "2",
"ecli": "ECLI:CE:ECHR:2025:1215JUD003161818"
}
```
### `paragraphs.json`
Array of paragraph objects:
```json
[
{"index": 0, "tag": "p", "text": "...", "char_count": 134},
...
]
```
### `task<N>/final.json`
Task-specific schemas. See the project repository for details.
## Source
Judgments retrieved from HUDOC — the official ECHR case-law database. Original
judgment texts are public-domain works of the European Court of Human Rights.
## License
MIT — annotations and processing code.
ECHR judgments themselves are public-domain works of the European Court of
Human Rights.
## Acknowledgements
Annotation produced via a Claude-assisted pipeline (Anthropic) with human review.
|