Bapynshngain/Bapyn-En-Kha

Dataset Description

Bapyn-En-Kha-Ext is an extended English–Khasi parallel corpus consisting of 70,065 aligned sentence pairs. It is designed to support research and development in:

• Neural Machine Translation (NMT)

• Cross-lingual representation learning

• Low-resource language modeling

• Multilingual AI applications involving Khasi and English

The dataset builds upon previously curated material and newly added sentence pairs to improve linguistic coverage, stylistic diversity, and sentence complexity.

Khasi is a low-resource Austroasiatic language primarily spoken in Meghalaya, India. Publicly available, high-quality parallel datasets for Khasi remain extremely limited. This corpus aims to reduce that gap.

Dataset Structure

Each record consists of two aligned fields:

en: English sentence

kha: Khasi sentence

Example

 {
  "en": "The quick brown fox jumps over the lazy dog.",
  "kha": "U myrsiang uba rong jngut u kynthih nalor u ksew ba jaipdeh."
 }

The dataset is provided in CSV format with two columns: en and kha.

Dataset Size

Total sentence pairs: 70,065

Languages: English ↔ Khasi

Domain: Mixed-domain (general conversation, narrative, descriptive, functional text)

Intended Uses

This dataset is suitable for:

• Training and fine-tuning English–Khasi translation models

• Building Khasi language tools (tokenizers, spell-checkers, embedding models)

• Multilingual NLP benchmarking

• Educational and academic research on low-resource languages

Limitations

Being a low-resource language, some stylistic and syntactic patterns may not fully cover all dialectal or cultural variations of Khasi. Some sentences may exhibit simplified constructions to preserve alignment quality. The dataset is not specifically domain-balanced (e.g., legal, medical, poetic domains may be underrepresented).

License

This dataset is released under the:

Creative Commons Attribution-NonCommercial 4.0 International (CC BY-NC 4.0)

You are free to:

• Share — copy and redistribute the material

• Adapt — remix, transform, and build upon the material

Under the following terms:

• Attribution required

• Non-commercial use only

• Citation

If you use this dataset in your research, please cite it as:

@dataset{bapyn_en_kha_ext,
  title   = {Bapyn-En-Kha-Ext: An Extended English-Khasi Parallel Corpus},
  author  = {Bapynshngainlang Nongkynrih},
  year    = {2025},
  license = {CC BY-NC 4.0},
  url     = {https://huggingface.co/datasets/Bapynshngain/Bapyn-En-Kha-Ext}
}

Inference:

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

model_name = "your-username/your-en-kha-model"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

text = "She couldn't decide if the glass was half empty or half full, so she drank it."

inputs = tokenizer(text, return_tensors="pt", padding=True, truncation=True)
outputs = model.generate(**inputs, max_length=128)

translation = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("Translated:", translation)

Creator

Bapynshngainlang Nongkynrih, Independent AI-Researcher