AfriSUD

AfriSUD is a collection of manually annotated treebanks for African languages following the Surface-Syntactic Universal Dependencies (SUD) framework. The dataset is intended to support research on dependency parsing, multilingual and cross-lingual NLP, and language technologies for low-resource African languages.

AfriSUD is introduced in our paper, AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages.

Loading the Dataset

Each language is available as a separate configuration.

For example, Hausa can be loaded with:

from datasets import load_dataset

dataset = load_dataset(
    "Lugha-Models/AfriSUD",
    "hau"
)

Languages

AfriSUD currently includes treebank data for eight African languages:

Dataset Format

The canonical release is provided in CoNLL-U format, with Parquet versions included for convenient access through the Hugging Face Dataset Viewer.

Annotation Guidelines

The annotation guidelines used to create AfriSUD are available here:

📄 SUD Annotation Guidelines

Paper

For more details about AfriSUD, including the annotation process, dataset construction, and evaluation experiments, see our paper:

AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages

📄 Paper on arXiv

Citation

If you use AfriSUD in your research, please cite:

@article{buzaaba2026afrisud,
  title={AfriSUD: A Dependency Treebank Collection for Evaluating Models on African Languages},
  author={Buzaaba, Happy and Dione, Cheikh Mouhamadou Bamba and Adelani, David Ifeoluwa and Kahane, Sylvain and Gerdes, Kim and Guillaume, Bruno and Muhammad, Shamsuddeen Hassan and Guan, Kevin and Anouluwapo, Aremu and Ettori, Naome A. and Inyang, Utitofon and Nabende, Peter and Bamutura, David Sabiiti and Bukula, Andiswa and Uchechukwu, Chinedu and Mabuya, Rooweither and Akinade, Idris and Fellbaum, Christiane},
  journal={arXiv preprint arXiv:2606.12708},
  year={2026}
}

Acknowledgments

This work was supported by Princeton Language and Intelligence (PLI).