AfriSpeech/afri-adaptations 🌍

Afri-Adaptations is a dataset of words that have been phonetically adapted into 723 African languages from English and French.

Every pair documents how foreign words, named entities, personal names, geographical places, and loanwords from English and French undergo phonological, morphological, and orthographic adaptation when borrowed into indigenous African languages. The dataset includes a universalised column generated using africa-g2p to provide cross-lingual phonetic grounding.

All pairs are extracted directly from authentic multi-source parallel text (AfriSpeech/africa-corpus and AfriSpeech/african-corpus-jw) using dependency-parsed noun-phrase alignments and consensus majority voting (100% human bitext, zero synthetic LLM translations).

• Author: Mich-Seth Owusu (michseth@ghananlp.org)
• Organization: AfriSpeech & Ghana NLP
• Codebase: github.com/AfriSpeech/afrigloss-mt
• African Languages Covered: 723 languages
• Total Directional Subsets: 1446 language pairs
  • English Source (en-*): 723 African target languages (868,684 adaptations)
  • French Source (fr-*): 723 African target languages (666,510 adaptations)
• Total Verified Word Adaptations: 1,535,194

---

📊 Sample Directional Subsets

Subset (config)	Language Direction	Verified Adapted Words
en-tw	English $\rightarrow$ tw	18,909
en-sw	English $\rightarrow$ sw	12,281
en-yo	English $\rightarrow$ yo	9,316
en-ha	English $\rightarrow$ ha	4,679
en-ig	English $\rightarrow$ ig	9,171
en-ee	English $\rightarrow$ ee	8,720
en-sn	English $\rightarrow$ sn	7,976
en-amh	English $\rightarrow$ amh	8,914
en-lin	English $\rightarrow$ lin	5,481
en-zul	English $\rightarrow$ zul	6,545
en-kin	English $\rightarrow$ kin	8,463
fr-tw	French $\rightarrow$ tw	12,591
fr-sw	French $\rightarrow$ sw	10,772
fr-yo	French $\rightarrow$ yo	8,789
fr-ha	French $\rightarrow$ ha	4,057
fr-ig	French $\rightarrow$ ig	7,766
fr-ee	French $\rightarrow$ ee	7,614
fr-sn	French $\rightarrow$ sn	7,066
fr-kin	French $\rightarrow$ kin	7,180
fr-loz	French $\rightarrow$ loz	5,432
...	(and 1426 more language directions)	...
Grand Total	1446 Language Pairs	1,535,194

---

🚀 How to Load Specific Language Directions

Load any specific directional subset directly with Hugging Face datasets:

from datasets import load_dataset

# 1. Load English -> Twi (Akan) phonetic adaptations
ds_en_tw = load_dataset("AfriSpeech/afri-adaptations", "en-tw", split="train")
print(ds_en_tw[0])
# {'source_term': 'Jehovah', 'target_term': 'Yehowa', 'universalised': 'yehowa', 'confidence': 0.9949, 'frequency': 32545}

# 2. Load English -> Swahili phonetic adaptations
ds_en_sw = load_dataset("AfriSpeech/afri-adaptations", "en-sw", split="train")
print(ds_en_sw[0])
# {'source_term': 'Jesus', 'target_term': 'Yesu', 'universalised': 'yesu', 'confidence': 0.9963, 'frequency': 18384}

# 3. Load English -> Amharic phonetic adaptations
ds_en_amh = load_dataset("AfriSpeech/afri-adaptations", "en-amh", split="train")
print(ds_en_amh[0])
# {'source_term': 'Jesus', 'target_term': 'ኢየሱስ', 'universalised': 'iyasuse', 'confidence': 1.0, 'frequency': 21841}

# 4. Load French -> Swahili phonetic adaptations
ds_fr_sw = load_dataset("AfriSpeech/afri-adaptations", "fr-sw", split="train")
print(ds_fr_sw[0])
# {'source_term': 'Jéhovah', 'target_term': 'Yehova', 'universalised': 'yehova', 'confidence': 0.9925, 'frequency': 45080}

Data Fields

• source_term: The original foreign word or named entity in English or French.
• target_term: The phonetically and orthographically adapted word in the African language.
• universalised: The phonetically universalized representation generated via africa-g2p, mapping African orthographies to common cross-lingual phonetic units.
• confidence: Empirical consensus score across all parallel sentence occurrences (0.00 – 1.00).
• frequency: Total number of occurrences observed across the corpus bitext.

---

⚖️ License

This dataset is distributed under the Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC 4.0) license.

---

📚 Citation

@article{owusu2026afrigloss,
  title={Afri-Gloss-MT: In-Place Glossary Constraints for African Machine Translation},
  author={Owusu, Mich-Seth},
  journal={arXiv preprint},
  year={2026}
}