Teklia/DAI-CReTDHI-RecordGold-ATR

---
dataset_info:
features:
- name: split
dtype: string
- name: source
dtype: string
- name: record_id
dtype: string
- name: record_url
dtype: string
- name: start_date
dtype: int64
- name: end_date
dtype: int64
- name: parish
dtype: string
- name: text
dtype: string
splits:
- name: train
num_bytes: 4229311
num_examples: 6178
- name: val
num_bytes: 531628
num_examples: 784
- name: test
num_bytes: 509437
num_examples: 758
download_size: 1908903
dataset_size: 5270376
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: val
path: data/val-*
- split: test
path: data/test-*
language:
- fr
tags:
- atr
- htr
- ocr
- historical
- handwritten
annotations_creators:
- expert-generated
license: mit
task_categories:
- image-to-text
pretty_name: DAI-records-atr
---

# DAI-records-ATR - Record level

## Dataset Description

- **Homepage:** [DAI-CReTDHI](https://dai-cretdhi.univ-lr.fr/)
- **Point of Contact:** [TEKLIA](https://teklia.com)

This dataset comprises parish and civil records from three main locations in France:
- [Ardennes](https://archives.cd08.fr/archives-numerisees)
- [Tours](https://archives.touraine.fr/)
- [Ile de Ré](https://archives.charente-maritime.fr/)

## Dataset Summary

The **DAI-records-ATR** dataset includes 7,720 handwritten records from the XVI-XIXth centuries.

The records have been annotated by experts as part of the [DAI-CReTDHI](https://dai-cretdhi.univ-lr.fr/) research project, using Teklia's open-source annotation interface [Callico](https://doc.callico.eu/).

### Split

| set | images |
| ----- | ------:|
| train | 6178 |
| val | 784 |
| test | 758 |

### Languages

All the documents in the dataset are written in French.

## Dataset Structure

### Data Instances

Each instance represents a record with its image and transcription:
```json
{
"split": "test",
"source": "Ardennes",
"record_id": "076e0576-7938-4d03-8e9d-0344d588ad92",
"record_url": "https://europe.iiif.teklia.com/iiif/2/geneanet%2FArdennes_BMS%2F382444%2F00104.jpg/1389,93,1234,443/full/0/default.jpg",
"start_date": 1629,
"end_date": 1696,
"parish": "Mézières",
"text": "1664\nLe samedi vingthuit jour de juin 1664 a esté baptisé Christophle du Raulin fils de Paulin Petitcolas et de Fabian Pessegniet ses pere et mere duquel ont esté parin et marine Christophle du Verdelet et Jeanne Petitcolas sa femme."}

```

### Data Fields

- `split` (string): Dataset split identifier (train, val, or test)
- `source` (string): Source collection
- `record_id` (string): Unique UUID identifier for the record in [Arkindex](https://arkindex.teklia.com/)
- `record_url` (string): IIIF URL to the image
- `start_date` (int): Start year of the corresponding parish register
- `end_date` (int): End year of the corresponding parish register
- `parish` (string): Name of the parish where the record originates
- `text` (`string`): Expert transcription. The transcription may include the following special tokens to indicate annotation uncertainty or editorial features:
- `[UNCERTAIN]`: Indicates a segment that could not be confidently transcribed by the annotator. This token replaces the unreadable or ambiguous portion.
- `[CROSSED_OUT]`: Placed immediately before a word or sequence of words that are crossed out in the original document. The crossed-out content itself is still transcribed and follows this token.