A cleaned and structured Nepali-language Bible dataset derived from the New Pioneers International Old and New Covenant Bible (NPIONCB) translation. This corpus is intended for use in pretraining foundational Nepali language models and other NLP research on low-resource Devanagari-script languages.

Dataset Summary

The Nepali Bible is a rich, formally structured source of written Nepali in Devanagari script. This dataset contains the full text of the Bible organized by book and chapter, cleaned and annotated with metadata for easy use in LLM training pipelines.

Property	Value
Language	Nepali (`ne`)
Script	Devanagari
Source	NPIONCB Nepali Bible
Format	JSONL / Parquet
Total chapters	~1,189
Splits	`train`, `validation`
Annotation	Metadata only (no manual labels)

Dataset Structure

Data Fields

Field	Type	Description
`text`	`string`	Full Nepali text of the chapter
`lang`	`string`	Language code — always `"ne"`
`source`	`string`	Always `"Nepali Bible (NPIONCB)"`
`book`	`string`	Bible book code (e.g. `GEN`, `MAT`, `REV`)
`chapter`	`int`	Chapter number within the book

Data Splits

Split	Records
`train`	~1,130
`validation`	~59

Sample

{
  "text": "आदिमा परमेश्‍वरले आकाशमण्डल र पृथ्वी सृष्‍टि गर्नुभयो। पृथ्वी आकारविनाको र शून्य थियो। गहिराइको सतहमाथि अन्धकार थियो, र पानीमाथि परमेश्‍वरको आत्मा घुमिरहनुहुन्थ्यो।",
  "lang": "ne",
  "source": "Nepali Bible (NPIONCB)",
  "book": "GEN",
  "chapter": 1
}

Usage

Load the full dataset

from datasets import load_dataset

ds = load_dataset("your-username/nepali-bible-corpus")
print(ds)
# DatasetDict({
#     train: Dataset({features: ['text', 'lang', 'source', 'book', 'chapter'], num_rows: 1130}),
#     validation: Dataset({features: ['text', 'lang', 'source', 'book', 'chapter'], num_rows: 59})
# })

Load only the training split

train_ds = load_dataset("your-username/nepali-bible-corpus", split="train")

Filter by book

genesis = ds["train"].filter(lambda x: x["book"] == "GEN")
new_testament = ds["train"].filter(lambda x: x["book"] in ["MAT", "MRK", "LUK", "JHN"])

Use as a text iterator for tokenizer training

def get_training_corpus():
    for example in ds["train"]:
        yield example["text"]

# Example: train a SentencePiece tokenizer
from tokenizers import SentencePieceBPETokenizer

tokenizer = SentencePieceBPETokenizer()
tokenizer.train_from_iterator(
    get_training_corpus(),
    vocab_size=32000,
    special_tokens=["<s>", "</s>", "<pad>", "<unk>"]
)

Intended Uses

✅ Suitable for

Pretraining foundational Nepali language models
Tokenizer training for Devanagari script
Benchmarking Nepali NLP models
Transfer learning for low-resource Nepali tasks
Linguistic research on formal written Nepali

❌ Not suitable for

Colloquial or spoken Nepali tasks (register mismatch)
Tasks requiring contemporary news or social media language
Direct use as a fine-tuning instruction dataset without further processing

Bible Book Codes Reference

Code	Book	Testament
GEN	Genesis	Old
EXO	Exodus	Old
PSA	Psalms	Old
PRO	Proverbs	Old
ISA	Isaiah	Old
MAT	Matthew	New
MRK	Mark	New
LUK	Luke	New
JHN	John	New
ACT	Acts	New
ROM	Romans	New
REV	Revelation	New

Dataset Creation

Source Data

The source text is the NPIONCB (New Pioneers International Old and New Covenant Bible) — a Nepali translation of the Bible in Devanagari script.

Processing Steps

Raw .txt files merged by book/chapter
Section headers (===== ... =====) removed
BOM characters (\ufeff) stripped
Bare verse/chapter number lines removed
Lines joined into full chapter-level text
Metadata (book, chapter, lang, source) extracted from filenames
95/5 train/validation split with seed=42

Who created this dataset?

This dataset was processed and uploaded as part of the foundational Nepali LLM project. The underlying Bible translation belongs to its respective copyright holders — please see the License section below.

Limitations & Biases

Register: The text is formal/religious Nepali. It does not represent colloquial, conversational, or modern everyday Nepali.
Vocabulary: Heavy use of theological and archaic vocabulary that may not appear in general-purpose Nepali text.
Size: ~1,189 chapters is a small corpus. For a robust foundational LLM, this should be combined with other Nepali corpora (Wikipedia, news, literature).
No annotation: No POS tags, NER labels, or syntactic parses are included.

Recommended Companion Datasets

To build a well-rounded foundational Nepali LLM, combine this dataset with:

Dataset	Source	Notes
Nepali Wikipedia	Wikimedia	Encyclopedic text
OSCAR Nepali	HuggingFace	Web-crawled text
CC-100 Nepali	HuggingFace	CommonCrawl Nepali
Nepali News Corpus	Various	News domain

from datasets import load_dataset, interleave_datasets

bible     = load_dataset("your-username/nepali-bible-corpus", split="train")
wikipedia = load_dataset("wikipedia", "20231101.ne", split="train")
cc100     = load_dataset("cc100", lang="ne", split="train")

# Interleave with sampling probabilities
combined = interleave_datasets(
    [bible, wikipedia, cc100],
    probabilities=[0.1, 0.4, 0.5],
    seed=42
)

Citation

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

@dataset{nepali_bible_npioncb_2024,
  author    = {Your Name / Your Organization},
  title     = {Nepali Bible Corpus (NPIONCB) for LLM Pretraining},
  year      = {2024},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/your-username/nepali-bible-corpus},
  language  = {Nepali},
  note      = {Derived from the NPIONCB Nepali Bible translation}
}

License

⚠️ Important: Please verify the specific license of the NPIONCB translation before public distribution. Bible translations vary — some are open (Creative Commons), others restrict commercial use.

This dataset card and processing code are released under CC BY 4.0. The underlying Bible text is subject to the license of the NPIONCB translation holders.

Contact

For questions about this dataset, open an issue on the dataset repository or contact the dataset maintainer.

Downloads last month: 9

Total file size:

3.69 MB