Hugging Face's logo

Datasets:
z00logist
/
OldChurchSlavonic

Tasks:
Text Generation
Modalities:
Text
Formats:
parquet
Languages:
Old Church Slavonic
Size:
100K - 1M
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
OldChurchSlavonic / README.md
z00logist's picture
z00logist
Add sources
27dc51b verified
preview code
|
raw
history blame contribute delete
5.91 kB
metadata 
license: cc-by-4.0
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
dataset_info:
  features:
    - name: Text
      dtype: string
    - name: Source
      dtype: string
  splits:
    - name: train
      num_bytes: 50647970
      num_examples: 233591
  download_size: 22877071
  dataset_size: 50647970
task_categories:
  - text-generation
language:
  - cu
size_categories:
  - 100K<n<1M

Old Church Slavonic Dataset

The Old Church Slavonic Dataset is a collection of preprocessed texts sourced from a combination of publicly accessible resources on the Internet and curated manuscripts from academic datasets. The dataset consists of structured texts from various Old Church Slavonic manuscripts, divided by sources (e.g., Akafist, Triodpost, Evangelie, etc.).

This dataset is designed to address the lack of dedicated datasets for Old Church Slavonic on Hugging Face. It preserves the rich linguistic features of Old Church Slavonic and provides a foundation for solving different NLP tasks with it.

This dataset was created with the goal of preserving and promoting the rich cultural and linguistic heritage of Slavic culture, making it accessible to modern computational tools. The aim is to encourage the exploration of Old Church Slavonic in scientific research, language modeling, and educational contexts.

Why This Dataset?

Old Church Slavonic is a liturgical language of Slavic cultures which now lacks dedicated machine learning datasets for text generation and related NLP tasks. Current resources on Hugging Face are scattered across different sources.

This dataset aims to:

  • Serve as a foundation for creating synthetic Old Church Slavonic text and parallel Ru-OCS datasets for research and experimentation.
  • Advance the understanding and appreciation of Old Church Slavonic as the first written Slavic language, foundational to Slavic culture and history.
  • Enable researchers, linguists, and developers to work on NLP tasks, such as text generation, translation, and analysis.
  • Support interdisciplinary studies in computational linguistics, history, and theology.

Dataset Structure

The dataset consists of preprocessed sentences divided by their sources. Each entry in the dataset represents a cleaned sample derived from a single file. The dataset uses the following format:

Column Description
text The preprocessed sentence from the source
source The source (e.g., "akafist", "bible")

Sources

The dataset contains texts from two main categories:

  1. Manuscripts from Lindstedt's Corpus Cyrillo-Methodianum Helsingiense
    A significant portion of the dataset includes texts derived from the following Old Church Slavonic manuscripts, curated in Lindstedt, J.'s Corpus Cyrillo-Methodianum Helsingiense:

    • Zographensis
    • Marianus
    • Assemanianus
    • Vitaconstantini
    • Savvina
    • Vitamethod
    • Zogrbb
    • Suprasliensis

    Citation for these manuscripts:

    Lindstedt, J. (2011). Corpus Cyrillo-Methodianum Helsingiense: Corpus of Old Church Slavonic Texts, source [data set]. Kielipankki. Retrieved from kielipankki.fi

  2. Texts from ortholib
    Other parts of the dataset include publicly available texts sourced from the http://www.orthlib.info/, such as:

    • Biblia: Old Church Slavonic translations of the Bible.
    • Akafist: Prayers and hymns to saints or events.
    • Triodpost: Lenten liturgical texts.
    • Evangelie: Gospel readings and manuscripts.
    • Psaltir: Psalms and accompanying liturgical hymns.
    • ... (and more, see the dataset structure on Hugging Face).

The author permits to distribute freely old church slavonic texts from the website. If you believe any content infringes on your rights, please contact me via provided info in my profile, including a clear statement of your claim. Once verified, I will quickly remove the disputed material.

  1. Hagiographies from the SCAT Project This portion includes 10 hagiographies of saints, sourced from the Saint Petersburg Corpus of Hagiographic Texts (SCAT):

    • Aleksandr Svirskij
    • Antonij Sijskij
    • Cyrill Novoezerskij
    • Dimitrij Prilutskij
    • Dionisij Glushitskij
    • Grigorij Pel'shemskij
    • Kornilij Komelskij
    • Pavel Obnorskij (zhitie and chudesa)
    • Sergij Nuromskij

    Citation for these hagiographies:

    Azarova, I. V., & Grebenkov, A. S. Saint Petersburg Corpus of Hagiographic Texts. Retrieved from https://project.phil.spbu.ru/scat/.

Preprocessing

  1. Removing Unwanted Sections:
    • Headers like "Часть," "Глава," and markers like "Кѡндакъ."
    • Verse markers, brackets, and other inline annotations.
  2. Normalizing Text:
    • Spaces normalized, punctuation preserved.
  3. Removing All-Caps Headers:
    • Eliminated uppercase section titles (common in manuscripts).
  4. Dividing by Files:
    • Each text file is treated as a single sample, with its full content preserved.

Intended Uses

This dataset is suitable for:

  • Text Generation Tasks: Fine-tuning or pretraining language models to generate Old Church Slavonic text.
  • Linguistic Research: Studying the structure and usage of Old Church Slavonic.
  • Synthetic Data Creation: Generating artificial text samples to augment the dataset or train specialized models.

Future iterations of this dataset will include synthetic texts generated from fine-tuned language models, further enriching the dataset for downstream tasks.

Limitations

  • Bias in Sources: The dataset reflects the content primarily focused on liturgical and church texts.
  • Limited Annotation: This dataset does not include linguistic or grammatical annotations.