Browse Datasets

MasakhaNER 2.0

Named Entity Recognition for 21 African Languages

named-entity-recognition sequence-labeling token-classification 50MB CC-BY-4.0
Access Dataset Read Paper

Description

MasakhaNER 2.0 is a comprehensive named entity recognition dataset covering 21 African languages. It includes annotations for Person (PER), Organization (ORG), Location (LOC), and Date (DATE) entities. Created by the Masakhane community, it represents a significant resource for African language NLP research.

Quick Start

Python 
from datasets import load_dataset

# Load Swahili subset
dataset = load_dataset("masakhane/masakhaner2", "swa")

# View a sample
print(dataset['train'][0])
# Output: {'tokens': ['Waziri', 'wa', ...], 'ner_tags': [1, 0, ...]}

# Get label names
print(dataset['train'].features['ner_tags'].feature.names)

Project Ideas

Multilingual News NER System

Build a system that extracts named entities from African news articles across multiple languages

Intermediate

Cross-lingual Transfer Study

Investigate how NER knowledge transfers between related African languages

Advanced

Ethical Considerations

  • Geographic bias towards regions with more annotators
  • Entity categories based on Western NER conventions may not cover all local contexts
  • Performance varies significantly across languages

Additional Information

MasakhaNER 2.0 builds on the original MasakhaNER dataset with improved coverage and annotation quality. The dataset was created through a collaborative effort by the Masakhane community, bringing together researchers and annotators from across Africa.

Supported Languages

The dataset covers 21 African languages:

  • East African: Swahili, Amharic, Luo, Luganda, Kinyarwanda
  • West African: Yoruba, Hausa, Igbo, Twi, Wolof, Bambara, Fon, Ewe, Mossi
  • Southern African: Zulu, Xhosa, Setswana, Shona, chiShona

Citation

@inproceedings{adelani-etal-2022-masakhaner,
    title = "{M}asakha{NER} 2.0: {A}frica-centric Transfer Learning for Named Entity Recognition",
    author = "Adelani, David and others",
    booktitle = "EMNLP",
    year = "2022"
}