Hugging Face's logo

Datasets:
TatarNLPWorld
/
tatar-english-russian-corpus

Tasks:
Translation
Text Generation
Modalities:
Tabular
Text
Formats:
parquet
Languages:
Tatar
Russian
English
Size:
10K - 100K
Tags:
tatar
russian
english
translate
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

Dataset Card: Tatar-English-Russian Parallel Corpus

Dataset Details

Dataset Description

This dataset is a parallel corpus containing 14,983 sentences in three languages: Tatar, English, and Russian. It combines two distinct sources:

  1. KickItLikeShika/english-tatar-translation (7,746 entries) – an existing English-Tatar dataset with Russian translations added
  2. yasalma/tt-en-language-corpus (7,615 entries) – another English-Tatar corpus with newly added Russian translations

The corpus is designed for machine translation, cross-lingual learning, and multilingual NLP tasks involving Tatar, a Turkic language spoken primarily in Tatarstan, Russia.

  • Curated by: TatarNLPWorld
  • Language(s): English (en), Tatar (tt), Russian (ru)
  • License: CC BY-SA 4.0
  • Size: 14,983 parallel sentences
  • Version: 1.0 (March 31, 2026)

Dataset Sources

Uses

Direct Use

This dataset can be used for:

  • Machine Translation: Training models for English↔Tatar, English↔Russian, or Tatar↔Russian translation
  • Multilingual NLP: Cross-lingual transfer learning and multilingual language modeling
  • Linguistic Research: Analyzing lexical and syntactic patterns across Turkic, Slavic, and Germanic languages
  • Educational Resources: Teaching materials for Tatar language learning
  • Benchmarking: Evaluating translation quality for low-resource languages (Tatar)

Out-of-Scope Use

The dataset should not be used for:

  • Evaluating model performance on formal documents (the corpus contains general phrases)
  • Tasks requiring domain-specific terminology (the dataset covers general vocabulary)
  • Sensitive applications without additional bias and safety evaluations

Dataset Structure

Data Fields

Field Type Description
english_text string English sentence
tatar_text string Tatar translation
russian_text string Russian translation
eng_len int64 Length (number of characters) of the English sentence
tat_len int64 Length (number of characters) of the Tatar sentence
rus_len int64 Length (number of characters) of the Russian sentence

Data Splits

Split Examples Description
full 14,983 Complete dataset
sample 1,000 Random sample for quick testing

Statistics

Overall:

  • Total records: 14,983
  • Empty English entries: 0 (0%)
  • Empty Tatar entries: 0 (0%)
  • Empty Russian entries: 57 (0.38%)

Text Length (characters):

Language Average Length
English 56.0
Tatar 57.9
Russian 58.4

Dataset Creation

Curation Rationale

This corpus was created to address the lack of multilingual parallel data for Tatar, a low-resource Turkic language. By combining two English-Tatar datasets with Russian translations, we provide a resource that enables tri-lingual machine translation and cross-lingual learning between Turkic, Slavic, and Germanic language families.

Source Data

Data Collection and Processing

The dataset was created by merging two independent English-Tatar corpora and adding Russian translations:

Source 1: KickItLikeShika/english-tatar-translation (7,746 entries)

  • Original dataset containing English-Tatar sentence pairs
  • Russian translations were added using Google Translate API
  • Processing details:
    • Library: googletrans
    • Checkpoint system saved progress every 10 sentences
    • Error handling with automatic retries
    • Output format: english_text, tatar_text, russian_text

Source 2: yasalma/tt-en-language-corpus (7,615 entries)

  • Another English-Tatar parallel corpus from Hugging Face
  • Russian translations added using the same Google Translate API workflow
  • Same processing pipeline with checkpointing and error recovery
  • Output format consistent with Source 1

Merging Process:

  1. Both datasets loaded as JSONL files (7,746 + 7,615 = 15,361 total)
  2. Combined into a single list
  3. Duplicate entries removed based on english_text field
  4. 378 duplicate entries removed
  5. Final dataset: 14,983 unique parallel sentences

Translation Implementation:

translator = Translator()
for i, english in enumerate(english_texts):
    try:
        result = translator.translate(english, src='en', dest='ru')
        russian = result.text
    except Exception as e:
        russian = ""
        time.sleep(2)  # Delay on error

Personal and Sensitive Information

The dataset contains general phrases and sentences with no personally identifiable information. It is suitable for public release under CC BY-SA 4.0 license.

Bias, Risks, and Limitations

  • Machine Translation Quality: Russian translations were generated using Google Translate API and may contain inaccuracies or unnatural phrasing. The translations were not manually verified.
  • Domain Coverage: The dataset consists of general phrases and may not cover specialized domains or formal language.
  • Tatar Language Variation: Tatar text may include variations in spelling and orthography across the two source datasets.
  • Balanced Representation: The corpus includes both simple and complex sentences but may not fully represent all linguistic phenomena.
  • Empty Russian Entries: 57 entries have empty Russian text (0.38%), which should be handled appropriately in downstream applications.
  • API Limitations: Translations were performed using a third-party API, which may have introduced inconsistencies due to rate limiting or service interruptions.

Recommendations

Users should:

  • Handle empty Russian entries when training models (e.g., filtering or masking)
  • Consider the machine-translated nature of Russian text for evaluation
  • Use the sample split for development and full split for final training
  • Augment with domain-specific data if targeting specialized applications
  • Be aware of potential inconsistencies between the two source datasets

Citation

If you use this dataset, please cite:

@misc{tatar-english-russian-corpus-2026,
  author = {Arabov Mullosharaf Kurbonovich, TatarNLPWorld},
  title = {Tatar-English-Russian Parallel Corpus},
  year = {2026},
  publisher = {Hugging Face},
  url = {https://huggingface.co/datasets/TatarNLPWorld/tatar-english-russian-corpus},
  note = {Combined dataset from KickItLikeShika/english-tatar-translation and yasalma/tt-en-language-corpus with Russian translations from Google Translate API}
}

License

This dataset is licensed under Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0).

Acknowledgments

This dataset was created by combining two sources with Russian translations:

Source Dataset Count Processing
Source 1 KickItLikeShika/english-tatar-translation 7,746 Russian translations added via Google Translate API
Source 2 yasalma/tt-en-language-corpus 7,615 Russian translations added via Google Translate API
Duplicates Removed Exact duplicates by English text 378 -
Total Final corpus 14,983 -

Tools used:

  • Google Translate API via googletrans for Russian translations
  • Hugging Face datasets library for data loading
  • Custom checkpoint system for reliable translation with error recovery (checkpoint every 10 sentences)
  • JSONL format for efficient storage and loading

Curation: TatarNLPWorld

Technical Details

Translation Process

  • Library: googletrans
  • Checkpointing: Progress saved every 10 sentences to checkpoint.txt
  • Error handling: Automatic retry with 2-second delay on errors
  • Total processing time: ~2.5 hours for 7,615 sentences (yasalma dataset)

Dataset Processing

  • Deduplication: 378 exact duplicates removed based on English text
  • Format: JSONL with UTF-8 encoding
  • Fields preserved from original sources with consistent naming

Dataset Card Authors

TatarNLPWorld

Dataset Card Contact

For questions or feedback, please open an issue on the Hugging Face repository.
Downloads last month
22
Total file size:
2.85 MB