README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: int64
    - name: domain
      dtype: string
    - name: source_lang
      dtype: string
    - name: target_lang
      dtype: string
    - name: source_text
      dtype: string
    - name: target_text
      dtype: string
  splits:
    - name: train
      num_bytes: 22209
      num_examples: 261
  download_size: 12982
  dataset_size: 22209
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-4.0
language:
  - en
  - nup
pretty_name: NupePilot
multilinguality: bilingual
size_categories:
  - n<1K
source_datasets:
  - original
task_categories:
  - text-generation
  - translation

NupePilot: A Multi-Domain English–Nupe Parallel Dataset

Dataset Description

Overview

NupePilot is a pilot parallel dataset for Nupe, a low-resource Niger-Congo language (Volta-Niger subfamily) spoken primarily in North-Central Nigeria.

Despite millions of speakers, Nupe remains significantly underrepresented in natural language processing (NLP), with only a small number of scattered datasets and limited structured resources available.

NupePilot provides a manually curated, multi-domain English–Nupe parallel corpus designed to support low-resource NLP research and applications.

---

Supported Tasks

• Machine Translation (English → Nupe)
• Text-to-Text Generation
• Conversational AI
• Cross-lingual transfer learning

---

Languages

• English (en)
• Nupe (nup)

---

Dataset Structure

Data Instances

Each example consists of:

{
  "id": 1,
  "domain": "conversation",
  "source_lang": "en",
  "target_lang": "nupe",
  "source_text": "What are you doing?",
  "target_text": "Ki wo-jon?"
}

---

Data Fields

• id: Unique identifier
• domain: Domain category (conversation, health, news)
• source_lang: Source language (English)
• target_lang: Target language (Nupe)
• source_text: Original sentence
• target_text: Translated sentence

Dataset Size

• Total examples: ~200+

Current Version Scope

This version of the dataset primarily contains everyday conversational phrases, with a smaller number of examples from additional domains such as:

• Health and public information
• News

Future versions of NupePilot will expand coverage to include more diverse domains, enabling broader applicability for NLP tasks.

Data Source

Sentences were curated from:

• Public-domain text sources
• Benchmark-style datasets (e.g., conversational and news corpora)
• Manually constructed examples
• Contemporary informational content

Personal and Sensitive Information

This dataset does not contain any personal or sensitive data.

Motivation

While recent years have seen progress in African NLP, many languages remain underrepresented. Nupe is one such example, with minimal digital presence and very few standardised datasets for machine learning.

This dataset is motivated by the need to:

• Enable machine translation for Nupe
• Support inclusive and equitable AI development
• Provide foundational data for future research
• Encourage community-driven language resource creation

Potential Impact

The rapid growth of AI technologies has enabled applications such as:

• Localised educational tools
• Language translation systems
• Conversational agents
• Healthcare information access

By open-sourcing this dataset, we aim to ensure that Nupe-speaking communities are not excluded from these advancements.

This dataset can serve as a foundation for building:

• Translation systems
• Chatbots/Conversational AI systems
• Language learning tools, and
• Public health communication systems for Nupe speakers

Considerations for Using the Data

Intended Use

This dataset is intended for:

• Academic research
• Low-resource NLP experimentation
• Prototyping translation systems
• Educational and linguistic analysis

Limitations

• Small dataset size (pilot-scale)
• Limited domain coverage

Biases

• Domain imbalance (limited domains, and currently highly skewed to everyday conservational sentences)
• Translation variability

Risks

• Not suitable for production-level systems
• May not generalise beyond included domains

Additional Information

Dataset Curators

• Amina Mardiyyah Rufai
• Fatima Tasallah Rufai
• Yahaya Gana Rufai

License

This dataset is released under the CC BY 4.0 License.

Credits

Created using Adaptive Data by Adaption.

Citation

If you use this dataset, please cite:

@dataset{nupepilot2026,
  title={NupePilot: A Multi-Domain English–Nupe Parallel Dataset},
  authors={amina mardiyyah rufai, fatima tasallah rufai},
  year={2026},
  url={https://huggingface.co/datasets/NupePilot/nupepilot}
}