agnivamaiti/nagamese-english-mt

Nagamese-English Machine Translation Corpus

Dataset Summary

This dataset contains 3,340 parallel sentence pairs in English and Nagamese (Naga Pidgin), intended for training and evaluating machine translation systems between the two languages. It has already been used to fine-tune at least one NLLB-200-based translation model (agnivamaiti/nllb-200-en-nagamese).

This is the first dataset card written for this dataset — no card existed on the repository prior to this version, so details below should be checked against your own records for accuracy (collection process, exact domains covered, etc.) before publishing.

Supported Tasks

• Machine Translation: English → Nagamese and Nagamese → English.

Languages

• English (en)
• Nagamese / Naga Pidgin (ISO 639-3: nag), a creole widely spoken across Nagaland and the wider Northeast India region.

Dataset Structure

Data Instance

{
  "english_sentence": "If the batting team loses early wickets in the powerplay, rebuilding the innings will require immense patience and skill.",
  "nagamese_sentence": "Judi batting team powerplay te joldi wicket khan harai jai, innings wapas bonabole dangor dhorjo aru gyaan lagibo."
}

Data Fields

• english_sentence: string — the English source sentence.
• nagamese_sentence: string — the corresponding Nagamese translation.

Data Splits

The dataset is split into three parts (train / validation / test — please confirm exact split names and sizes against the repository, as the split breakdown was not fully visible at card-writing time):

Split	Approx. rows
train	majority of 3,340
(other splits)	remainder

Total rows across all splits: 3,340.

Dataset Creation

Source Data

Sentence pairs span a range of registers and domains — everyday narration, sports commentary, descriptive prose, and dialogue-style sentences — designed to give translation models broad coverage rather than a single narrow domain.

Annotations

Nagamese translations were produced/reviewed as part of the author's Nagamese NLP work (the same NagaNLP research effort behind the other datasets in this collection).

Considerations for Using the Data

Limitations

• Moderate scale (3,340 pairs) — sufficient for fine-tuning a pretrained multilingual MT model (e.g., NLLB-200) via transfer learning, but not for training a translation model from scratch.

Licensing

Licensed under CC BY-NC 4.0 (attribution required, non-commercial use only). For commercial licensing inquiries, please contact the author.

Citation

If you use this dataset, please cite:

@misc{maiti2026nagamesemt,
      title={Nagamese-English Machine Translation Corpus},
      author={Agniva Maiti},
      year={2026},
      howpublished={\url{https://huggingface.co/datasets/agnivamaiti/nagamese-english-mt}}
}

Contact

Agniva Maiti — KIIT University — HuggingFace: agnivamaiti