README.md

metadata

license: apache-2.0
configs:
  - config_name: bam
    data_files:
      - split: robotsmali
        path: bam/robotsmali/*
      - split: dokotoro
        path: bam/dokotoro/*
      - split: flores
        path: bam/flores/*
      - split: mafand
        path: bam/mafand/*
      - split: xp3
        path: bam/xp3/*
      - split: opus
        path: bam/opus/*
      - split: bible
        path: bam/bible/*
      - split: sentiments
        path: bam/sentiments/*
      - split: donkibaru
        path: bam/donkibaru/*
      - split: corbama
        path: bam/corbama/*
      - split: smol
        path: bam/smol/*
      - split: rfi
        path: bam/rfi/*
      - split: jw
        path: bam/jw/*
      - split: fakan
        path: bam/fakan/*
      - split: constitution
        path: bam/constitution/*
  - config_name: taq
    data_files: []
  - config_name: bmq
    data_files: []
  - config_name: mku
    data_files: []
  - config_name: dtm
    data_files: []
  - config_name: myk
    data_files: []
  - config_name: boz
    data_files: []
  - config_name: ses
    data_files: []
  - config_name: ful
    data_files: []
  - config_name: snk
    data_files: []
  - config_name: mlq
    data_files: []
  - config_name: hsy
    data_files: []
  - config_name: seq
    data_files: []
task_categories:
  - feature-extraction
  - token-classification
  - text-classification
  - sentence-similarity
language:
  - bm
  - mey
  - taq
  - bmq
  - myk
  - sep
  - boz
  - bze
  - khq
  - ses
  - mlq
  - mku
  - dtm
  - ff
pretty_name: sebenx
size_categories:
  - 1M<n<10M

West African Multilingual Tokenized Corpus

This repository contains a high-quality, linguistically annotated corpus for West African languages, featuring deep tokenization, lemmatization, and morphological analysis. The dataset is designed to support advanced NLP tasks such as POS tagging, lemmatization, and machine translation for under-resourced languages.

📊 Data Structure

Each entry in the dataset consists of a raw sentence and a highly detailed Tokens object. Unlike standard tokenizers, this dataset provides nested linguistic metadata for every word.

Column Definitions

• Sentence: The raw string of text in the target language.
• Tokens: A JSON object containing:
• text: Individual tokens (words and punctuation).
• start / end: Character offsets within the original sentence.
• type: Classification (e.g., word, character).
• lemmas: A list of potential lemmas, Parts of Speech (ps), French glosses (gloss_fr), and detailed morphemes breakdowns.

Data Example (Bamanankan)

Sentence: Kumaw yɛrɛ kɛcogo fɛ e n'a dɔn...

Token Breakdown (JSON snippet):

{
  "text": ["Kumaw", "yɛrɛ"],
  "lemmas": [
    {
      "lemma": ["kumaw"],
      "ps": ["n"],
      "gloss_fr": ["parole"],
      "morphemes": [{"morph": ["kúma", "w"], "ps": ["n", "mrph"], "gloss_fr": ["parole", "PL"]}
      ]
    }
  ]
}

---

📂 Configuration & Splits

The dataset is partitioned by ISO 639-3 language codes. The current primary focus is Bamanankan (bam), with multiple specialized sub-splits:

Config	Split Name	Description
bam	robotsmali	All major NLP related internal datasets
bam	corbama	Corpus Bamana de Référence
bam	donkibaru	News and media
bam	constitution	Legal and administrative
bam	flores / xp3	Benchmark and instruction sets

🌍 Language Coverage Roadmap

Currently Active / Populated

• Bamanankan (bam): Extensive coverage across 15+ sources.
• Initial Configs Ready: taq (Tamasheq), bmq (Bomu), mku (Konyanka), dtm (Tomo Kan), myk (Mamara), ses (Koyraboro Senni), ful (Fulfulde), snk (Soninke).

Future Coverage

We are expanding to cover over 100+ languages across the Sahel and West Africa, including:

Mali & Guinea

• Mali: Hasanya Arabic (mey), Maasina Fulfulde (ffm), various Dogon dialects (dds, dto, dts), and Bozo varieties.
• Guinea: Susu (sus), N’Ko (nqo), Pular (fuf), and Toma (tod).

Burkina Faso & Niger

• Burkina Faso: Moore (mos), Jula (dyu), Lobi (lob), and Lyélé (lee).
• Niger: Hausa (hau), Zarma (dje), and various Kanuri dialects.

Senegal & Cross-Border

• Senegal: Wolof (wol), Serer-Sine (srr), and Pulaar (fuc).
• Cross-Border: Tamasheq (taq), Mandinka (mnk), and Dazaga (dzg).

Below is the complete list of target languages categorized by region and ISO 639-3 code.

🤝 Shared & Cross-Border

Code	Language	Code	Language
beh	Biali	bof	Bolon
bmq	Bomu	fuh	Burkina Faso Fulfulde
dzg	Dazaga	ddn	Dendi
emk	Eastern Maninkakan	gux	Gourmanchéma
hmb	Humburi Senni Songhay	yal	Jalunga
djm	Jamsay Dogon	dyo	Jola-Fonyi
bbo	Konabéré	kfz	Koromfé
mnk	Mandinka	mfv	Mandjak
rkm	Marka	stj	Matya Samo
bsc	Oniyan	pnq	Pana
fuc	Pulaar	fuf	Pular
sym	Sàmòmá	srr	Serer-Sine
shu	Shuwa Arabic	sep	Sìcìté Sénoufo
snk	Soninke	ttq	Tamajaq
taq	Tamasheq	tuq	Tedaga
dtm	Tomo Kan Dogon	cou	Wamey
mlq	Western Maninkakan	knc	Yerwa Kanuri

🇲🇱 Mali

Code	Language	Code	Language
bam	Bamanankan	dbw	Bankan Tey Dogon
dbt	Ben Tey Dogon	dbu	Bondum Dom Dogon
dds	Donno So Dogon	mey	Hasanya Arabic
bze	Jenaama Bozo	bzx	Kelengaxo Bozo
mwk	Kita Maninkakan	khq	Koyra Chiini Songhay
ses	Koyraboro Senni Songhay	ffm	Maasina Fulfulde
myk	Mamara Sénoufo	seb	Shempire Sénoufo
spp	Supyire Sénoufo	shz	Syenara Sénoufo
dsq	Tadaksahak	boo	Tiemacèwè Bozo
boz	Tieyaxo Bozo	dtk	Togo Kan Dogon
dto	Tommo So Dogon	dts	Toro So Dogon
dtt	Toro Tegu Dogon	kao	Xaasongaxango

🇧🇫 Burkina Faso

Code	Language	Code	Language
mos	Moore	bib	Bisa
box	Buamu	cme	Cerma
bwy	Cwi Bwamu	dgd	Dagaari Dioula
dos	Dogosé	dya	Dyan
dnn	Dzùùngoo	kar	Eastern Karaboro
dyu	Jula	gna	Kaansa
xsm	Kasem	cpo	Kpeego
kus	Kusaal	bwj	Láá Láá Bwamu
lob	Lobi	lee	Lyélé
bfo	Malba Birifor	sym	Maya Samo
mfq	Moba	sen	Nanerigé Sénoufo
gur	Ninkare	dgi	Northern Dagara
nuv	Northern Nuni	tsp	Northern Toussian
pug	Phuie	sos	Seenku
seq	Senara Sénoufo	sif	Siamou
sld	Sissala	bwq	Southern Bobo Madaré
nnw	Southern Nuni	sbd	Southern Samo
wib	Southern Toussian	lor	Téén
tiq	Tiéfo	tuz	Turka
kza	Western Karaboro	kst	Winyé

🇬🇳 Guinea

Code	Language	Code	Language
bqf	Baga Kaloum	bsp	Baga Sitemu
dnj	Dan	lma	East Limba
gkp	Guinea Kpelle	jad	Jahanka
kke	Kakabe	lda	Kla-Dan
kno	Kono	mku	Konyanka
knk	Kuranko	ldm	Landoma
llc	Lele	mev	Mano
mzj	Manya	mfg	Mogofin
nqo	N’Ko	naj	Nalu
kqs	Northern Kissi	msc	Sankaran Maninka
sus	Susu	tod	Toma
zil	Zialo

🇳🇪 Niger

Code	Language	Code	Language
dje	Zarma	bms	Bilma Kanuri
fuq	Central-Eastern Niger Fulfulde	hau	Hausa
ayl	Libyan Arabic	kby	Manga Kanuri
tda	Tagdal	thv	Tahaggart Tamahaq
twq	Tasawaq	thz	Tayart Tamajeq
tez	Tetserret	krt	Tumari Kanuri

🇸🇳 Senegal

Code	Language	Code	Language
bbg	Bainouk-Gunyaamolo	bcb	Bainouk-Samik
bjt	Balanta-Ganja	bqj	Bandial
bda	Bayot	eja	Ejamat
gsl	Gusilay	csk	Jola-Kasa
krx	Karon	cwt	Kuwaataay
cae	Laalaa	knf	Mankanya
ndv	Ndut	snf	Noon
fap	Paloor	sav	Saafi-Saafi
wol	Wolof

---

🛠 Usage

You can load the data using the Hugging Face datasets library:

from datasets import load_dataset

# Load Bamanankan news data
ds = load_dataset("s7d11/sebenx", "bam", split="donkibaru")

# Access deep tokenization
first_row = ds[0]
print(f"Sentence: {first_row['Sentence']}")
print(f"Tokens: {first_row['Tokens']['text']}")

---

📜 License

This dataset is released under the Apache-2.0 License.

---