0xmezbah/bangla-literary-corpus

Bangla Literary Corpus - বাংলা সাহিত্য সংকলন

Bangla Literary Corpus is a large-scale, chapter-segmented Bengali literature dataset with over 3,200 books including classical novels (উপন্যাস), short stories (ছোটগল্প), essays (প্রবন্ধ), and poetry collections (কবিতা). Curated specifically for Bengali NLP and Indic language modeling, this corpus provides high-register, linguistically diverse Bangla text optimized for large language model (LLM) pretraining, continual pre-training (CPT), tokenizer vocabulary training, stylometry, and retrieval-augmented generation (RAG).

All texts are conservatively cleaned, chapter-indexed with source provenance, and distributed in sharded, Snappy-compressed Apache Parquet format for zero-overhead streaming and fast distributed training.

Dataset Summary

Metric	Value
Total Books	3,261
Total Chapters	78,278
Total Characters	1,179,518,078 (~1.18 Billion)
Distinct Authors	792
Curated Series	43
Distinct Genres	30
Language	Bengali (bn)
Format	Apache Parquet (Snappy-compressed)
Script Verification	>80% verified Bengali Unicode per document

Quick Start

1. Load via Hugging Face datasets

from datasets import load_dataset

chapters = load_dataset("0xmezbah/bangla-literary-corpus", "chapters", split="train")
print(chapters[0])

# Load full-length books (recommended for long-context LLMs >=32k)
books = load_dataset("0xmezbah/bangla-literary-corpus", "books", split="train")
print(books[0])

2. Streaming

from datasets import load_dataset

stream = load_dataset("0xmezbah/bangla-literary-corpus", "chapters", split="train", streaming=True)
for sample in stream:
    print(sample["book_title"], "—", sample["chapter_title"])
    break

3. Read directly with Polars or Pandas

import polars as pl

df = pl.read_parquet("data/chapters/train-00000-of-00040.parquet")
print(df.head())

Schemas

chapters

One record per chapter or sub-section.

Field	Type	Description
id	string	Unique chapter identifier (<book_id>#<index>)
book_id	string	Foreign key referencing the parent book
book_title	string	Full title of the book
author	string	Standardized author name
genres	list[string]	Assigned genre tags
series	string / null	Series or omnibus name (if applicable)
chapter_index	int64	1-based sequential position in the book
chapter_title	string	Title of the chapter or poem
level	string	Structural level (lesson for chapter, topic for poem/sub-chapter)
chars	int64	Character count of the text
source_url	string	Canonical source URL for provenance and auditing
text	string	Verbatim chapter body

books

One record per entire book.

Field	Type	Description
id	string	Unique book slug
title	string	Book title as published
author	string	Standardized author name
genres	list[string]	Assigned genre tags
series	string / null	Associated series name
n_chapters	int64	Total chapters compiled into this work
chars	int64	Total character count of full book
chapter_titles	list[string]	Ordered list of all chapter headers
source_url	string	Canonical source URL
text	string	Complete book text (chapters delimited by \n\n)

Corpus Distribution

Top Genres by Title Count

Genre	Books	Genre	Books
বাংলা উপন্যাস	999	আত্মজীবনী ও স্মৃতিকথা	134
অনুবাদ বই	271	গোয়েন্দা-ডিটেকটিভ	122
প্রবন্ধ ও গবেষণা	270	ইতিহাস ও সংস্কৃতি	90
বাংলা গল্প	237	ভৌতিক	72
থ্রিলার / রহস্য	196	ইসলামিক সাহিত্য	64
কিশোর সাহিত্য	160	রচনাবলী / সংকলন	62
বাংলা কবিতা	156	ধর্ম ও দর্শন	53
সেবা প্রকাশনী	149	সায়েন্স ফিকশন	46

Prominent Authors

Author	Books	Author	Books
হুমায়ূন আহমেদ	197	আশাপূর্ণা দেবী	39
রবীন্দ্রনাথ ঠাকুর	119	সমরেশ মজুমদার	36
সুনীল গঙ্গোপাধ্যায়	71	শরৎচন্দ্র চট্টোপাধ্যায়	29
সমরেশ বসু	59	বঙ্কিমচন্দ্র চট্টোপাধ্যায়	28
মুহম্মদ জাফর ইকবাল	58	সৈয়দ মুজতবা আলী	28
বুদ্ধদেব গুহ	53	বিভূতিভূষণ বন্দ্যোপাধ্যায়	22
কাজী আনোয়ার হোসেন	51	হুমায়ুন আজাদ	22
শীর্ষেন্দু মুখোপাধ্যায়	47	মানিক বন্দ্যোপাধ্যায়	21
সত্যজিৎ রায়	45	জীবনানন্দ দাশ	18
কাজী নজরুল ইসলাম	40	মাইকেল মধুসূদন দত্ত	14

Companion Catalog (catalog/)

The repository includes metadata tables in CSV/TXT format inside the catalog/ directory:

• catalog.csv: Master index with metadata, URLs, character counts, and chapter counts.
• authors.csv: Author rollups with book counts, total character volume, and genre associations.
• series.csv: Index of all 43 literary series and franchises.
• genres.csv: Breakdown of titles across all 30 genre taxonomies.
• authors.txt & titles.txt: Flat newline-delimited lists for easy indexing and lookups.

Terms of Use & Fair Dealing Notice

This corpus is distributed strictly for educational, non-commercial linguistic research, academic benchmarking, and NLP model training.

• Ownership: No copyright ownership is claimed over the literary contents. All proprietary rights remain with their respective authors, estates, and publishers.
• Legal Context: Provided in accordance with Fair Dealing principles under Section 72 of the Bangladesh Copyright Act 2000 and Section 107 of the U.S. Copyright Act (Title 17).
• Takedown Requests: If you are a copyright owner or authorized representative and request the exclusion of a specific text, please open a discussion ticket for prompt removal.

Citation

If you utilize this corpus in your research or applications, please cite it as:

@misc{bangla_literary_corpus_2026,
  author = {MD Mezbah Ul Islam},
  title = {Bangla Literary Corpus: A Large-Scale Open Dataset of Bengali Literature},
  year = {2026},
  publisher = {Hugging Face},
  journal = {Hugging Face Hub},
  howpublished = {\url{https://huggingface.co/datasets/0xmezbah/bangla-literary-corpus}}
}