Dataset Viewer

The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider removing the loading script and relying on automated data support (you can use convert_to_parquet from the datasets library). If this is not possible, please open a discussion for direct help.

Dataset Card for MasakhaNER

Dataset Summary

MasakhaNER-X is an aggregation of MasakhaNER 1.0 and MasakhaNER 2.0 datasets for 20 African languages. The dataset is not in CoNLL format. The input is the original raw text while the output is byte-level span annotations.

Example: {"example_id": "test-00015916", "language": "pcm", "text": "By Bashir Ibrahim Hassan", "spans": [{"start_byte": 3, "limit_byte": 24, "label": "PER"}], "target": "PER: Bashir Ibrahim Hassan"}

MasakhaNER-X is a named entity dataset consisting of PER, ORG, LOC, and DATE entities annotated by Masakhane for twenty African languages:

  • Amharic
  • Ghomala
  • Bambara
  • Ewe
  • Hausa
  • Igbo
  • Kinyarwanda
  • Luganda
  • Luo
  • Mossi
  • Chichewa
  • Nigerian-Pidgin
  • chiShona
  • Swahili
  • Setswana
  • Twi
  • Wolof
  • Xhosa
  • Yoruba
  • Zulu

The train/validation/test sets are available for all the twenty languages.

For more details see https://aclanthology.org/2022.emnlp-main.298

Supported Tasks and Leaderboards

[More Information Needed]

  • named-entity-recognition: The performance in this task is measured with Span F1 (higher is better). A named entity is correct only if it is an exact match of the corresponding entity in the data.

Languages

There are twenty languages available :

  • Amharic (am)
  • Ghomala (bbj)
  • Bambara (bm)
  • Ewe (ee)
  • Hausa (ha)
  • Igbo (ig)
  • Kinyarwanda (rw)
  • Luganda (lg)
  • Luo (luo)
  • Mossi (mos)
  • Chichewa (ny)
  • Nigerian-Pidgin (pcm)
  • chiShona (sn)
  • Swahili (sw)
  • Setswana (tn)
  • Twi (tw)
  • Wolof (wo)
  • Xhosa (xh)
  • Yoruba (yo)
  • Zulu (zu)

Dataset Structure

Data Instances

The examples look like this for Nigerian-Pidgin:

from datasets import load_dataset
data = load_dataset('masakhaner-x', 'pcm') 

# Please, specify the language code

# A data point consists of sentences seperated by empty line and tab-seperated tokens and tags. 
{'id': '0',
 'text': "Most of de people who dey opposed to Prez Akufo-Addo en decision say within 3 weeks of lockdown, total number of cases for Ghana rise from around 100 catch 1024.",
 'spans': [{"start_byte": 42, "limit_byte": 52, "label": "PER"}, {"start_byte": 76, "limit_byte": 83, "label": "DATE"}, {"start_byte": 123, "limit_byte": 128, "label": "LOC"}]
 'target': "PER: Akufo-Addo $$ DATE: 3 weeks $$ LOC: Ghana"
}

Data Fields

  • id: id of the sample
  • text: sentence containing entities
  • spans: details of each named entities in the sentence
  • target: named entities and their values. Each named entity is separated by '$$'

The NER tags correspond to this list:

"PER", "ORG", "LOC", and "DATE",

Data Splits

For all languages, there are three splits - train, validation and test splits.

The splits have the following sizes :

Language train validation test
Amharic 1441 250 500
Gbomola 1441 483 966
Bambara 1441 638 1000
Ewe 1441 501 1000
Hausa 1441 1000 1000
Igbo 1441 319 638
Kinyarwanda 1441 1000 1000
Luganda 1441 906 1000
Luo 644 92 185
Mossi 1441 648 1000
Chichewa 1441 893 1000
Nigerian-Pidgin 1441 1000 1000
Shona 1441 887 1000
Swahili 1441 1000 1000
Setswana 1441 499 996
Twi 1441 605 1000
Wolof 1441 923 1000
Xhosa 1441 817 1000
Yoruba 1441 1000 1000
Zulu 1441 836 1000

Licensing Information

The licensing status of the data is CC 4.0 Non-Commercial

Citation Information

Provide the BibTex-formatted reference for the dataset. For example:

@inproceedings{adelani-etal-2022-masakhaner,
    title = "{M}asakha{NER} 2.0: {A}frica-centric Transfer Learning for Named Entity Recognition",
    author = "Adelani, David  and
      Neubig, Graham  and
      Ruder, Sebastian  and
      Rijhwani, Shruti  and
      Beukman, Michael  and
      Palen-Michel, Chester  and
      Lignos, Constantine  and
      Alabi, Jesujoba  and
      Muhammad, Shamsuddeen  and
      Nabende, Peter  and
      Dione, Cheikh M. Bamba  and
      Bukula, Andiswa  and
      Mabuya, Rooweither  and
      Dossou, Bonaventure F. P.  and
      Sibanda, Blessing  and
      Buzaaba, Happy  and
      Mukiibi, Jonathan  and
      Kalipe, Godson  and
      Mbaye, Derguene  and
      Taylor, Amelia  and
      Kabore, Fatoumata  and
      Emezue, Chris Chinenye  and
      Aremu, Anuoluwapo  and
      Ogayo, Perez  and
      Gitau, Catherine  and
      Munkoh-Buabeng, Edwin  and
      Memdjokam Koagne, Victoire  and
      Tapo, Allahsera Auguste  and
      Macucwa, Tebogo  and
      Marivate, Vukosi  and
      Elvis, Mboning Tchiaze  and
      Gwadabe, Tajuddeen  and
      Adewumi, Tosin  and
      Ahia, Orevaoghene  and
      Nakatumba-Nabende, Joyce  and
      Mokono, Neo Lerato  and
      Ezeani, Ignatius  and
      Chukwuneke, Chiamaka  and
      Oluwaseun Adeyemi, Mofetoluwa  and
      Hacheme, Gilles Quentin  and
      Abdulmumin, Idris  and
      Ogundepo, Odunayo  and
      Yousuf, Oreen  and
      Moteu, Tatiana  and
      Klakow, Dietrich",
    booktitle = "Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing",
    month = dec,
    year = "2022",
    address = "Abu Dhabi, United Arab Emirates",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.emnlp-main.298",
    pages = "4488--4508",
    abstract = "African languages are spoken by over a billion people, but they are under-represented in NLP research and development. Multiple challenges exist, including the limited availability of annotated training and evaluation datasets as well as the lack of understanding of which settings, languages, and recently proposed methods like cross-lingual transfer will be effective. In this paper, we aim to move towards solutions for these challenges, focusing on the task of named entity recognition (NER). We present the creation of the largest to-date human-annotated NER dataset for 20 African languages. We study the behaviour of state-of-the-art cross-lingual transfer methods in an Africa-centric setting, empirically demonstrating that the choice of source transfer language significantly affects performance. While much previous work defaults to using English as the source language, our results show that choosing the best transfer language improves zero-shot F1 scores by an average of 14{\%} over 20 languages as compared to using English.",
}
Downloads last month
9

Collection including masakhane/masakhaner-x