failed09/bashkir-books-kitap

Bashkir Books & Calendars Cleaned (Kitap)

Curated, cleaned and structured Bashkir literature and annual calendars (Kitap Publishing House) for LLM pretraining and fine-tuning.

Overview

A curated, deduplicated and filtered edition of modern Bashkir book publications and annual cultural calendars derived from the original bashkorttele/books-kitap dataset published by the Foundation for the Preservation and Development of the Bashkir Language.

The dataset includes material from 36 books and annual calendars published by the Zainab Biisheva Bashkir Publishing House "Kitap" (including folklore tales «Әкиәт иленә сәйәхәт», prose by Farzana Akbulatova, Aibulat Sasanbayev, Ralif Kinyabaev, and 12 annual wall/desk cultural calendars spanning 2014–2026). Repetitive astronomical tables (sunrise/sunset, moon phases, Hijri dates), catalog metadata (УДК, ББК, ISBN), and publisher boilerplate were systematically removed, and language purity was verified using neural language identification.

At a glance
Task	Text generation, Fill-mask, Language Modeling
Default config	sections (6,044 sections/entries, 1.26M words)
Fields	id, book_id, segmentation, text, n_words, n_chars, n_sentences, ba_prob, source_id
Source	Zainab Biisheva Bashkir Publishing House "Kitap" (via bashkorttele/books-kitap)
License	CC-BY-4.0

Contents

Files and Configurations

Configuration	File	Rows	Words	Description
sections (default)	cleaned_sections.parquet	5,641	1,132,418	Coherent daily calendar articles and book chapters
sentences	cleaned_sentences.parquet	73,821	647,205	Deduplicated sentences (3..40 words, NER-protected, strict syntax)

Schema

sections configuration:

Column	Type	Description
id	string	Unique record identifier (e.g. kitap-2016-tp-164-15-block-002)
book_id	string	Publication identifier (e.g. kitap-2016-tp-164-15, kitap-si-2026-skazki)
segmentation	string	Source segmentation type (days, headings, toc, whole)
text	string	Cleaned section/article text
n_words	int64	Word count of the cleaned text
n_chars	int64	Character count of the cleaned text
n_sentences	int64	Number of extracted sentences
ba_prob	float64	Neural Bashkir LID confidence score (0.90..1.00)
source_id	string	Upstream block identifier

sentences configuration:

Column	Type	Description
sentence_id	string	Unique sentence identifier
text	string	Clean sentence text
book_id	string	Publication identifier
word_count	int64	Sentence word count
source_id	string	Upstream block identifier

Examples

Sections configuration:

Book ID	Preview	ba_prob	Words
kitap-2016-tp-164-15	50 йыл элек (1966, 1 ғинуар) Өфө телеүҙәгенән СССР-ҙың Үҙәк телевидениеһы программаларының даими трансляцияһы башлана...	1.0000	258
kitap-si-2026-skazki	АЙЫУ МЕНӘН БАЛ ҠОРТТАРЫ орон-борон заманда, өйрәк үрәтник, һайыҫҡан дисәтник саҡта булған, ти, бер тайыш табан айыу...	1.0000	114

Sentences configuration:

Book ID	Text	Words
kitap-2016-tp-164-15	һәр кемдең ҡалалыр бер йәдкәре...	5
kitap-2016-tp-164-15	Йыл иртәһе тиҙәр... ғинуарҙы, Тимәк, ғинуар йылды башланы.	8
kitap-2017-tp-144-16	Күтәреләйек беҙ үҙебеҙ!	3

Method

upstream jsonl (6,081 records)
  ├── 1. strip calendar astronomical headers (sunrise, moon phases, hijri dates)
  ├── 2. strip bibliographic catalog lines (УДК, ББК, ISBN, publisher colophons)
  ├── 3. reject syllabified primers & multi-column interlaced corrupted scans (405 dropped)
  ├── 4. normalize whitespace and detached OCR punctuation ("һүҙ ." -> "һүҙ.")
  ├── 5. anonymize publisher contacts: mask phones ([ТЕЛЕФОН]) and emails ([EMAIL])
  ├── 6. neural language filter: failed09/bashkir-lid (ba_prob >= 0.90) (35 non-ba dropped)
  ├── 7. sentence segmentation with initials/abbreviation protection, casing & syntax filters, deduplication (73,821 unique sentences)
  └── 8. export zstd Parquet: sections (5,641) & sentences (73,821)

Quality and Use

Data Sanitization & PII Protection

To protect personal privacy, all texts were passed through automated PII detection. Phone numbers and publisher email addresses have been masked with standard placeholders: [ТЕЛЕФОН] and [EMAIL].

Limitations

• Fully Automated Pipeline: Text extraction, boilerplate stripping, language identification, and sentence splitting were executed entirely automatically. Consequently, residual OCR artifacts from upstream PDF digitization, broken hyphenations, rare fragmented phrases, or imperfect sentence boundary detections may remain.
• Folklore and Literary Vocabulary: Texts contain rich idiomatic, dialectal, and traditional vocabulary (especially in folk tales and author collections), representing literary and historical registers of Bashkir.
• Calendar Origins: Daily calendar entries represent concise biographies, cultural essays, culinary recipes, and folk wisdom, distinct from long-form novels.

Related Resources

• Bashkir LID (Binary BA vs NON_BA) — compact ONNX language gate used for neural language filtering (ba_prob >= 0.90) in this dataset.
• Bashkir-Russian Phrasebook Corpus — reviewed and cleaned conversational parallel corpus, also derived and optimized from bashkorttele releases.
• Bashkir Periodicals Cleaned (Izddom) — companion dataset containing 38,039 cleaned articles and 1.66M sentences from Bashkir magazines and newspapers.

Usage

pip install datasets pyarrow

from datasets import load_dataset

# Load full sections/stories for pretraining
dataset = load_dataset("failed09/bashkir-books-kitap", "sections", split="train")
print(dataset[0]["text"][:200])

# Load deduplicated sentences
sentences = load_dataset("failed09/bashkir-books-kitap", "sentences", split="train")
print(sentences[0])

Or directly with Pandas:

import pandas as pd

df = pd.read_parquet("cleaned_sections.parquet")
print(df.head())

License

Distributed under the CC BY 4.0 license. The release contains derived text extracted from books and calendars published by the Zainab Biisheva Bashkir Publishing House "Kitap" and processed by the Foundation for the Preservation and Development of the Bashkir Language (bashkorttele). Original rights remain with the authors and publishers; users should preserve upstream attribution.

Citation

@dataset{failed09_bashkir_books_kitap_2026,
  title = {Bashkir Books and Calendars Cleaned (Kitap)},
  author = {failed09},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/failed09/bashkir-books-kitap},
  note = {Derived from bashkorttele/books-kitap}
}

Open Bashkir Data and Sources 🐝

This release is part of an open-source effort to support the development, preservation and practical use of the Bashkir language. Other related models, datasets and tools are available on the author's Hugging Face profile.

The author does not claim ownership or authorship of the source texts or other materials used to derive this release; rights and licensing remain with the original authors, publishers and dataset providers. Source texts are not redistributed in this repository, so users should follow the licenses and attribution requirements of the relevant upstream resources.