Update README.md
Browse files
README.md
CHANGED
|
@@ -2,32 +2,37 @@
|
|
| 2 |
language:
|
| 3 |
- ff
|
| 4 |
- fr
|
| 5 |
-
- en
|
| 6 |
-
- ar
|
| 7 |
tags:
|
| 8 |
- pulaar
|
| 9 |
- fulfulde
|
| 10 |
- fulani
|
| 11 |
-
-
|
| 12 |
-
-
|
| 13 |
-
-
|
| 14 |
- NLP
|
| 15 |
- low-resource
|
|
|
|
| 16 |
license: cc-by-4.0
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 17 |
---
|
| 18 |
|
| 19 |
-
# ARPRIM —
|
| 20 |
-
##
|
| 21 |
|
| 22 |
-
**Projet ARPRIM** | pulaar.org
|
| 23 |
|
| 24 |
---
|
| 25 |
|
| 26 |
## Description
|
| 27 |
|
| 28 |
-
Ce dataset
|
| 29 |
|
| 30 |
-
|
| 31 |
|
| 32 |
---
|
| 33 |
|
|
@@ -35,148 +40,154 @@ Le Pulaar/Fulfulde est parlé par plus de 40 millions de locuteurs en Afrique de
|
|
| 35 |
|
| 36 |
| Indicateur | Valeur |
|
| 37 |
|---|---|
|
| 38 |
-
| Total
|
| 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 |
-
| elevage_botanique | 83 |
|
| 68 |
-
| adjectifs_etats | 81 |
|
| 69 |
-
| sciences_nature | 74 |
|
| 70 |
-
| alimentation_sante | 66 |
|
| 71 |
-
| reseaux_sociaux | 63 |
|
| 72 |
-
| informatique_logiciels | 56 |
|
| 73 |
-
| terminologie_ARPRIM | 37 |
|
| 74 |
-
| physique | 28 |
|
| 75 |
-
| pedagogie | 23 |
|
| 76 |
-
| couleurs_nature | 12 |
|
| 77 |
-
| sante_mentale_psychiatrie | 12 |
|
| 78 |
|
| 79 |
---
|
| 80 |
|
| 81 |
-
## Format des données (
|
|
|
|
|
|
|
| 82 |
|
| 83 |
```json
|
| 84 |
{
|
| 85 |
-
"id": "
|
| 86 |
-
"
|
| 87 |
-
"francais": "
|
| 88 |
-
"
|
| 89 |
-
"
|
| 90 |
-
"source": "
|
| 91 |
}
|
| 92 |
```
|
| 93 |
|
| 94 |
-
Champs disponibles :
|
| 95 |
-
- `id` : identifiant unique ARPRIM
|
| 96 |
-
- `domaine` : catégorie thématique
|
| 97 |
-
- `francais` : terme en français
|
| 98 |
-
- `pulaar` : terme en Pulaar/Fulfulde
|
| 99 |
-
- `anglais` : terme en anglais *(si disponible)*
|
| 100 |
-
- `arabe` : terme en arabe *(si disponible)*
|
| 101 |
-
- `pulaar_alt` : variante dialectale *(si disponible)*
|
| 102 |
-
- `note` : note ou contexte *(si disponible)*
|
| 103 |
-
- `source` : source du terme
|
| 104 |
-
|
| 105 |
---
|
| 106 |
|
| 107 |
-
##
|
| 108 |
|
| 109 |
-
|
| 110 |
-
|
| 111 |
-
|
| 112 |
-
|
| 113 |
-
|
| 114 |
-
|
| 115 |
-
|
| 116 |
-
|
| 117 |
-
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
| Doosgal_RIM_FBPM | — | 163 |
|
| 123 |
-
| MSM | — | 123 |
|
| 124 |
-
| HIISIWAL_GFW | — | 109 |
|
| 125 |
-
| ngaynaaka | — | 99 |
|
| 126 |
-
| G5-Sahel | — | 97 |
|
| 127 |
-
| elevage_ASB | — | 83 |
|
| 128 |
-
| sifaaji | — | 81 |
|
| 129 |
-
| Livre3-SN_ILN | — | 74 |
|
| 130 |
-
| alim | — | 66 |
|
| 131 |
-
| facebook | — | 63 |
|
| 132 |
-
| Abiword-IMS | — | 56 |
|
| 133 |
-
| ILN-ARPRIM | — | 37 |
|
| 134 |
-
| GIYL | — | 32 |
|
| 135 |
-
| balliwal_MFS | — | 28 |
|
| 136 |
-
| pedagogie | — | 23 |
|
| 137 |
-
| noordi | — | 12 |
|
| 138 |
-
| Sba-GATA | — | 12 |
|
| 139 |
|
| 140 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 141 |
|
| 142 |
-
##
|
|
|
|
|
|
|
|
|
|
|
|
|
| 143 |
|
| 144 |
-
|
| 145 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
|
| 147 |
---
|
| 148 |
|
| 149 |
-
##
|
|
|
|
|
|
|
| 150 |
|
| 151 |
-
|
| 152 |
-
import json
|
| 153 |
|
| 154 |
-
|
| 155 |
-
data = json.load(f)
|
| 156 |
|
| 157 |
-
#
|
| 158 |
-
tic = [e for e in data if e["domaine"] == "informatique_logiciels"]
|
| 159 |
|
| 160 |
-
|
| 161 |
-
|
| 162 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 163 |
|
| 164 |
---
|
| 165 |
|
| 166 |
## Licence
|
| 167 |
|
| 168 |
-
CC BY 4.0 —
|
|
|
|
| 169 |
|
| 170 |
---
|
| 171 |
|
| 172 |
## Citation
|
| 173 |
|
| 174 |
-
```
|
| 175 |
-
@dataset{
|
| 176 |
-
title={ARPRIM —
|
| 177 |
-
author={Projet ARPRIM},
|
| 178 |
-
year={2025},
|
| 179 |
-
|
| 180 |
-
|
|
|
|
|
|
|
| 181 |
}
|
| 182 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 2 |
language:
|
| 3 |
- ff
|
| 4 |
- fr
|
|
|
|
|
|
|
| 5 |
tags:
|
| 6 |
- pulaar
|
| 7 |
- fulfulde
|
| 8 |
- fulani
|
| 9 |
+
- parallel-corpus
|
| 10 |
+
- sentence-alignment
|
| 11 |
+
- health
|
| 12 |
- NLP
|
| 13 |
- low-resource
|
| 14 |
+
- translation
|
| 15 |
license: cc-by-4.0
|
| 16 |
+
task_categories:
|
| 17 |
+
- translation
|
| 18 |
+
- text-generation
|
| 19 |
+
pretty_name: ARPRIM — Corpus de phrases alignées Français/Pulaar (Santé & Société)
|
| 20 |
+
size_categories:
|
| 21 |
+
- n<1K
|
| 22 |
---
|
| 23 |
|
| 24 |
+
# ARPRIM — Corpus de phrases alignées Français / Pulaar
|
| 25 |
+
## Santé publique & Société — SOS Pairs Éducateurs
|
| 26 |
|
| 27 |
+
**Projet ARPRIM** | [pulaar.org](https://pulaar.org) | [HuggingFace/ARPRIM](https://huggingface.co/ARPRIM)
|
| 28 |
|
| 29 |
---
|
| 30 |
|
| 31 |
## Description
|
| 32 |
|
| 33 |
+
Ce dataset contient **232 paires de phrases alignées** Français ↔ Pulaar/Fulfulde, issues des scripts de sensibilisation de l'**Association SOS Pairs Éducateurs** (Mauritanie), produits en collaboration avec la Commune de N'Diago, BP et ses partenaires, World Vision, les Nations Unies et le Ministère de la Santé mauritanien.
|
| 34 |
|
| 35 |
+
Il s'agit de **phrases complètes, contextualisées**, couvrant des thèmes de santé publique et de cohésion sociale — un type de données rare et précieux pour les langues à faibles ressources comme le Pulaar/Fulfulde.
|
| 36 |
|
| 37 |
---
|
| 38 |
|
|
|
|
| 40 |
|
| 41 |
| Indicateur | Valeur |
|
| 42 |
|---|---|
|
| 43 |
+
| Total paires alignées | 232 |
|
| 44 |
+
| Langue source | Français |
|
| 45 |
+
| Langue cible | Pulaar / Fulfulde |
|
| 46 |
+
| Nombre de thèmes | 15 |
|
| 47 |
+
| Type de contenu | Phrases complètes (sensibilisation) |
|
| 48 |
+
| Longueur moyenne FR | ~120 caractères |
|
| 49 |
+
| Longueur moyenne Pulaar | ~130 caractères |
|
| 50 |
|
| 51 |
---
|
| 52 |
|
| 53 |
+
## Thèmes couverts
|
| 54 |
|
| 55 |
+
| Thème | Paires |
|
| 56 |
|---|---|
|
| 57 |
+
| VIH / Sida | 38 |
|
| 58 |
+
| Vaccination COVID-19 | 33 |
|
| 59 |
+
| Suivi de grossesse | 27 |
|
| 60 |
+
| Confinement COVID-19 | 25 |
|
| 61 |
+
| Lèpre (Baras) | 20 |
|
| 62 |
+
| Extrémisme et radicalisation | 20 |
|
| 63 |
+
| Autotest VIH | 14 |
|
| 64 |
+
| Grippe aviaire | 12 |
|
| 65 |
+
| Lutte contre la drogue | 10 |
|
| 66 |
+
| Discipline personnelle | 10 |
|
| 67 |
+
| Fièvre Vallée du Rift | 6 |
|
| 68 |
+
| Poliomyélite | 6 |
|
| 69 |
+
| Fièvre Crimée-Congo | 4 |
|
| 70 |
+
| Hépatite B | 4 |
|
| 71 |
+
| Général | 3 |
|
| 72 |
+
| **TOTAL** | **232** |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 73 |
|
| 74 |
---
|
| 75 |
|
| 76 |
+
## Format des données (JSONL)
|
| 77 |
+
|
| 78 |
+
Chaque ligne est un objet JSON avec les champs suivants :
|
| 79 |
|
| 80 |
```json
|
| 81 |
{
|
| 82 |
+
"id": "SOS_0001",
|
| 83 |
+
"theme": "Vaccination COVID-19",
|
| 84 |
+
"francais": "Durant les 12 derniers mois, nous avons tous vécu dans l'angoisse et la peur d'être contaminés par la Covid-19.",
|
| 85 |
+
"pulaar": "Lebbi 12 kuurɗi, eɗen ngondi, enen e wondiiɓe men, e kulol raaɓeede Cowid 19.",
|
| 86 |
+
"domaine": "sante_societe",
|
| 87 |
+
"source": "SOS_Pairs_Educateurs_Mauritanie"
|
| 88 |
}
|
| 89 |
```
|
| 90 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 91 |
---
|
| 92 |
|
| 93 |
+
## Usages possibles
|
| 94 |
|
| 95 |
+
### 1. Entraînement / fine-tuning de modèles de traduction
|
| 96 |
+
```python
|
| 97 |
+
from datasets import load_dataset
|
| 98 |
+
ds = load_dataset("ARPRIM/pulaar_fulfulde", data_files="SOS_PAIRS_paires.jsonl")
|
| 99 |
+
|
| 100 |
+
# Formater pour la traduction
|
| 101 |
+
def format_translation(example):
|
| 102 |
+
return {
|
| 103 |
+
"input": f"Traduire en Pulaar : {example['francais']}",
|
| 104 |
+
"output": example['pulaar']
|
| 105 |
+
}
|
| 106 |
+
ds = ds.map(format_translation)
|
| 107 |
+
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 108 |
|
| 109 |
+
### 2. Évaluation de modèles existants
|
| 110 |
+
```python
|
| 111 |
+
# Tester un modèle de traduction sur ce corpus
|
| 112 |
+
from transformers import pipeline
|
| 113 |
+
translator = pipeline("translation", model="Helsinki-NLP/opus-mt-fr-ff")
|
| 114 |
+
|
| 115 |
+
for pair in ds['train']:
|
| 116 |
+
pred = translator(pair['francais'])[0]['translation_text']
|
| 117 |
+
ref = pair['pulaar']
|
| 118 |
+
# calcul BLEU, chrF, etc.
|
| 119 |
+
```
|
| 120 |
|
| 121 |
+
### 3. Extraction de vocabulaire médical
|
| 122 |
+
```python
|
| 123 |
+
import json
|
| 124 |
+
with open("SOS_PAIRS_paires.json", encoding="utf-8") as f:
|
| 125 |
+
data = json.load(f)
|
| 126 |
|
| 127 |
+
# Filtrer par thème
|
| 128 |
+
sante = [p for p in data if p['theme'] in ['VIH / Sida', 'Hépatite B', 'Lèpre (Baras)']]
|
| 129 |
+
print(f"{len(sante)} phrases sur la santé")
|
| 130 |
+
```
|
| 131 |
+
|
| 132 |
+
### 4. Augmentation de données (back-translation)
|
| 133 |
+
Ce corpus peut servir de corpus de départ pour générer des données augmentées via back-translation FR→Pulaar→FR.
|
| 134 |
+
|
| 135 |
+
### 5. Intégration dans un système RAG (Retrieval-Augmented Generation)
|
| 136 |
+
Les paires alignées permettent de construire un index vectoriel pour un assistant de santé communautaire en Pulaar.
|
| 137 |
|
| 138 |
---
|
| 139 |
|
| 140 |
+
## Orthographe de référence
|
| 141 |
+
|
| 142 |
+
L'orthographe du Pulaar utilisée dans ce corpus suit la variante mauritanienne (**Pulaar de Mauritanie / Halpulaar**), proche de la norme de Bamako 2010 (ACALAN-UA).
|
| 143 |
|
| 144 |
+
Caractères spéciaux utilisés : **ɓ ɗ ƴ ñ ŋ ɗ** — encodés en UTF-8.
|
|
|
|
| 145 |
|
| 146 |
+
---
|
|
|
|
| 147 |
|
| 148 |
+
## Source et crédits
|
|
|
|
| 149 |
|
| 150 |
+
| Élément | Détail |
|
| 151 |
+
|---|---|
|
| 152 |
+
| Producteur des scripts | Association SOS Pairs Éducateurs (Mauritanie) |
|
| 153 |
+
| Partenaires | Commune de N'Diago, BP, World Vision, Nations Unies, Ministère de la Santé RIM |
|
| 154 |
+
| Compilation ARPRIM | Projet ARPRIM — pulaar.org |
|
| 155 |
+
| Langue source | Français |
|
| 156 |
+
| Langue cible | Pulaar / Fulfulde (variante mauritanienne) |
|
| 157 |
|
| 158 |
---
|
| 159 |
|
| 160 |
## Licence
|
| 161 |
|
| 162 |
+
**CC BY 4.0** — Utilisation libre avec attribution :
|
| 163 |
+
> *Projet ARPRIM / SOS Pairs Éducateurs Mauritanie — pulaar.org*
|
| 164 |
|
| 165 |
---
|
| 166 |
|
| 167 |
## Citation
|
| 168 |
|
| 169 |
+
```bibtex
|
| 170 |
+
@dataset{ARPRIM_SOSPairs_2025,
|
| 171 |
+
title = {ARPRIM — Corpus de phrases alignées Français/Pulaar : Santé \& Société},
|
| 172 |
+
author = {Projet ARPRIM and SOS Pairs Éducateurs},
|
| 173 |
+
year = {2025},
|
| 174 |
+
publisher = {HuggingFace},
|
| 175 |
+
url = {https://huggingface.co/datasets/ARPRIM/pulaar_fulfulde},
|
| 176 |
+
license = {cc-by-4.0},
|
| 177 |
+
note = {232 paires alignées, 15 thèmes de santé publique et société}
|
| 178 |
}
|
| 179 |
```
|
| 180 |
+
|
| 181 |
+
---
|
| 182 |
+
|
| 183 |
+
## Datasets connexes ARPRIM
|
| 184 |
+
|
| 185 |
+
| Dataset | Description | Entrées |
|
| 186 |
+
|---|---|---|
|
| 187 |
+
| ARPRIM/Pulaar_Dictionary | Terminologies multidomaines | 7 719 |
|
| 188 |
+
| ARPRIM/Pulaar_Fulfulde_Corpus | Corpus général | — |
|
| 189 |
+
| **Ce dataset** | Phrases alignées Santé & Société | 232 |
|
| 190 |
+
|
| 191 |
+
---
|
| 192 |
+
|
| 193 |
+
*Pulaar/Fulfulde — 40+ millions de locuteurs — Sénégal · Mauritanie · Mali · Guinée · Burkina Faso · Niger · Cameroun · Nigeria · Soudan*
|