README.md

---
license: other
language:
- en
- am
- ff
- yo
- ig
- om
- sw
- ha
- tw
- sn
- rw
- ee
- bm
- wo
- lg
- ar
- so
- af
- ti
- mg
- xh
- zu
- st
- ny
- ln
- tn
- fon
- ki
- lua
- mos
- kg
- bem
task_categories:
- translation
tags:
- african-languages
- machine-translation
- low-resource-languages
- arxiv:2510.05644
pretty_name: African Languages Lab Multi-Open
configs:
- config_name: english-amharic
  data_files:
  - split: train
    path: "data/english-amharic/train-*.parquet"
- config_name: english-fula
  data_files:
  - split: train
    path: "data/english-fula/train-*.parquet"
- config_name: english-yoruba
  data_files:
  - split: train
    path: "data/english-yoruba/train-*.parquet"
- config_name: english-igbo
  data_files:
  - split: train
    path: "data/english-igbo/train-*.parquet"
- config_name: english-oromo
  data_files:
  - split: train
    path: "data/english-oromo/train-*.parquet"
- config_name: english-swahili
  data_files:
  - split: train
    path: "data/english-swahili/train-*.parquet"
- config_name: english-hausa
  data_files:
  - split: train
    path: "data/english-hausa/train-*.parquet"
- config_name: english-twi
  data_files:
  - split: train
    path: "data/english-twi/train-*.parquet"
- config_name: english-shona
  data_files:
  - split: train
    path: "data/english-shona/train-*.parquet"
- config_name: english-kinyarwanda
  data_files:
  - split: train
    path: "data/english-kinyarwanda/train-*.parquet"
- config_name: english-ewe
  data_files:
  - split: train
    path: "data/english-ewe/train-*.parquet"
- config_name: english-bambara
  data_files:
  - split: train
    path: "data/english-bambara/train-*.parquet"
- config_name: english-wolof
  data_files:
  - split: train
    path: "data/english-wolof/train-*.parquet"
- config_name: english-luganda
  data_files:
  - split: train
    path: "data/english-luganda/train-*.parquet"
- config_name: english-arabic
  data_files:
  - split: train
    path: "data/english-arabic/train-*.parquet"
- config_name: english-somali
  data_files:
  - split: train
    path: "data/english-somali/train-*.parquet"
- config_name: english-afrikaans
  data_files:
  - split: train
    path: "data/english-afrikaans/train-*.parquet"
- config_name: english-tigrinya
  data_files:
  - split: train
    path: "data/english-tigrinya/train-*.parquet"
- config_name: english-malagasy
  data_files:
  - split: train
    path: "data/english-malagasy/train-*.parquet"
- config_name: english-xhosa
  data_files:
  - split: train
    path: "data/english-xhosa/train-*.parquet"
- config_name: english-zulu
  data_files:
  - split: train
    path: "data/english-zulu/train-*.parquet"
- config_name: english-southern sotho
  data_files:
  - split: train
    path: "data/english-southern_sotho/train-*.parquet"
- config_name: english-chewa
  data_files:
  - split: train
    path: "data/english-chewa/train-*.parquet"
- config_name: english-lingala
  data_files:
  - split: train
    path: "data/english-lingala/train-*.parquet"
- config_name: english-tswana
  data_files:
  - split: train
    path: "data/english-tswana/train-*.parquet"
- config_name: english-fon
  data_files:
  - split: train
    path: "data/english-fon/train-*.parquet"
- config_name: english-kikuyu
  data_files:
  - split: train
    path: "data/english-kikuyu/train-*.parquet"
- config_name: english-tshiluba
  data_files:
  - split: train
    path: "data/english-tshiluba/train-*.parquet"
- config_name: english-mossi
  data_files:
  - split: train
    path: "data/english-mossi/train-*.parquet"
- config_name: english-kikongo
  data_files:
  - split: train
    path: "data/english-kikongo/train-*.parquet"
- config_name: english-bemba
  data_files:
  - split: train
    path: "data/english-bemba/train-*.parquet"
---

# African Languages Lab Multi-Open

`multi-open` is the open-source multilingual subset released by the
[African Languages Lab](https://www.africanlanguageslab.com/). It contains English-target
parallel text for 31 African languages.

**Project website:** https://the-african-languages-lab.github.io/

**[The African Languages Lab: A Collaborative Approach to Advancing Low-Resource African
NLP](https://aclanthology.org/2026.acl-long.1965/)**  
Issaka et al., ACL 2026.

The paper presents All Lab's broader collaborative program: systematic and quality-controlled
data infrastructure, empirical evaluation for low-resource African NLP, and local research
capacity building. It reports a broader multimodal collection spanning 40 languages and evaluates
translation across 31 languages. **The full dataset is not released here.** The complete paper corpus is access-restricted and requires a license for full release.

## ACL 2026 oral presentation

This work is presented as an **oral paper at ACL 2026 (San Diego)**. The conference poster:

![The African Languages Lab — ACL 2026 poster](poster.jpg)

## Data format

Each configuration is a separate English-target parallel dataset. Text and token-count column
names follow the language pair (for example, `english`, `swahili`, `english_token_count`, and
`swahili_token_count`).

## Token counts

Token counts per language pair (source and target `*_token_count` columns).

| Language pair | Number of rows | English tokens | Target tokens | Combined tokens |
|---|--:|--:|--:|--:|
| english-arabic | 9,999,986 | 289,643,600 | 405,315,900 | 694,959,500 |
| english-amharic | 5,000,002 | 78,282,582 | 87,467,465 | 165,750,047 |
| english-afrikaans | 5,000,006 | 72,617,092 | 78,248,827 | 150,865,919 |
| english-swahili | 3,000,001 | 45,619,235 | 50,012,735 | 95,631,970 |
| english-yoruba | 850,000 | 26,153,024 | 40,944,851 | 67,097,875 |
| english-xhosa | 800,001 | 11,898,318 | 13,318,543 | 25,216,861 |
| english-somali | 599,999 | 8,938,590 | 14,680,753 | 23,619,343 |
| english-hausa | 449,997 | 10,440,301 | 11,727,876 | 22,168,177 |
| english-igbo | 350,002 | 8,301,596 | 10,741,795 | 19,043,391 |
| english-kinyarwanda | 500,001 | 7,140,011 | 8,916,183 | 16,056,194 |
| english-malagasy | 420,000 | 6,576,333 | 8,406,495 | 14,982,828 |
| english-tigrinya | 420,000 | 6,049,701 | 7,557,984 | 13,607,685 |
| english-luganda | 300,000 | 5,745,054 | 7,436,919 | 13,181,973 |
| english-oromo | 410,000 | 4,499,080 | 6,899,616 | 11,398,696 |
| english-kikuyu | 380,000 | 4,747,061 | 6,394,913 | 11,141,974 |
| english-wolof | 400,000 | 4,896,950 | 5,607,835 | 10,504,785 |
| english-tshiluba | 400,000 | 4,478,381 | 5,769,974 | 10,248,355 |
| english-zulu | 300,000 | 4,518,161 | 5,230,191 | 9,748,352 |
| english-fon | 270,000 | 3,738,341 | 5,973,240 | 9,711,581 |
| english-fula | 300,000 | 4,224,575 | 5,278,610 | 9,503,185 |
| english-lingala | 390,000 | 4,384,898 | 4,812,085 | 9,196,983 |
| english-shona | 300,000 | 4,196,647 | 4,779,603 | 8,976,250 |
| english-southern_sotho | 300,000 | 3,815,716 | 4,863,376 | 8,679,092 |
| english-chewa | 250,000 | 4,010,771 | 4,472,410 | 8,483,181 |
| english-twi | 300,000 | 3,701,483 | 4,249,403 | 7,950,886 |
| english-mossi | 300,000 | 3,372,164 | 4,410,104 | 7,782,268 |
| english-kikongo | 330,000 | 3,420,676 | 4,138,205 | 7,558,881 |
| english-tswana | 300,000 | 2,977,328 | 3,739,706 | 6,717,034 |
| english-bemba | 250,000 | 2,737,222 | 3,414,609 | 6,151,831 |
| english-bambara | 100,000 | 1,307,668 | 1,753,615 | 3,061,283 |
| english-ewe | 52,478 | 1,346,840 | 1,611,353 | 2,958,193 |
| **Total** | **33,022,473** | **643,779,399** | **828,175,174** | **1,471,954,573** |


## Usage

```python
from datasets import load_dataset

dataset = load_dataset("African-Languages-Lab/multi-open", "english-swahili")
```

## Citation

```bibtex
@inproceedings{issaka-etal-2026-african,
  title = "The {A}frican Languages Lab: A Collaborative Approach to Advancing Low-Resource {A}frican {NLP}",
  author = "Issaka, Sheriff and Wang, Keyi and Ajibola, Yinka and Samuel-Ipaye, Oluwatumininu and Zhang, Zhaoyi and Jimenez, Nicte Aguillon and Agyei, Evans Kofi and Lin, Abraham and Ramachandran, Rohan and Mumin, Sadick Abdul and Nchifor, Faith and Issah, Mohammed Shuraim and Gonzalez, Erick Rosas and Liu, Lieqi and Kpei, Sylvester and Osei, Jemimah Kusi and Ajeneza, Carlene and Boateng, Persis and Yeboah, Prisca Adwoa Dufie and Gabriel, Saadia",
  booktitle = "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
  month = jul,
  year = "2026",
  publisher = "Association for Computational Linguistics",
  url = "https://aclanthology.org/2026.acl-long.1965/",
  pages = "42460--42477"
}
```


## Note on translation_quality_score

`translation_quality_score` is an **automatic 1–10 translation-quality score** from a Gemma judge (`google/gemma-4-26B-A4B-it`), for filtering/analysis only — **not** human gold labels. **Null** values mean the judge output could not be parsed.