NHLOCAL/project-ben-yehuda

Project Ben-Yehuda Corpus

Dataset Summary

This dataset contains Hebrew literary texts sourced from Project Ben-Yehuda and structured for NLP, language modeling, text analysis, historical linguistics, and computational Hebrew research.

Each record represents one text/work and includes the full text, a fixed source label, and structured metadata from the Project Ben-Yehuda catalogue.

Current Version

Dataset version: pby-2026.03
Source snapshot: Project Ben-Yehuda, March 2026
Build date: 2026-05-04
Format: Parquet
Shards: 10
Expected rows: 26,455
Split: train

Versioning Policy

Versions use the following scheme:

pby-YYYY.MM[.patch]

• YYYY.MM refers to the Project Ben-Yehuda source snapshot month.
• .patch is used only for packaging, metadata, README, or shard-layout fixes that do not change the corpus content.
• Examples:
  • pby-2026.03 - March 2026 source snapshot.
  • pby-2026.03.1 - packaging or metadata fix for the same source snapshot.
  • pby-2026.06 - new source snapshot from June 2026.

Dataset Structure

The dataset has one split:

train

Each record contains:

• text: full text content.
• source: constant string, Project Ben-Yehuda.
• metadata: structured metadata object.

Metadata Fields

• pby_id: Project Ben-Yehuda catalogue ID.
• title: title of the work.
• authors: author name(s).
• translators: translator name(s), if applicable.
• author_uris: Wikidata URI(s) for author(s), when available.
• translator_uris: Wikidata URI(s) for translator(s), when available.
• original_language: original language for translated works, when available.
• genre: catalogue genre.
• source_edition: source edition information.
• filepath_pby: original Project Ben-Yehuda archive path.

Data Files

This release is distributed as 10 Parquet shards:

pby_dataset-part-00001.parquet
pby_dataset-part-00002.parquet
...
pby_dataset-part-00010.parquet

Loading the Dataset

from datasets import load_dataset

dataset = load_dataset("NHLOCAL/project-ben-yehuda")
train = dataset["train"]

print(train[0]["text"])
print(train[0]["metadata"])

Dataset Creation

The dataset was built by:

1. Reading pseudocatalogue.csv from the Project Ben-Yehuda source snapshot.
2. Matching catalogue rows to their corresponding .txt files.
3. Creating structured records with text, source, and metadata.
4. Writing directly to sharded Parquet files.
5. Splitting the output into 10 shards for easier upload, tracking, and downstream consumption.

The build scripts are maintained here:

https://github.com/NHLOCAL/The-Digital-Genizah

Changelog

pby-2026.03

• Rebuilt from Project Ben-Yehuda March 2026 source snapshot.
• Increased corpus coverage to 26,455 records.
• Replaced the previous single-file/JSONL-derived pipeline with direct Parquet generation.
• Published as 10 Parquet shards.
• Preserved schema: text, source, metadata.

Source Data and Attribution

This dataset is based on texts made available by Project Ben-Yehuda and its volunteers.

Project Ben-Yehuda website:

https://benyehuda.org/

This dataset is an independent packaging and processing effort. It is not officially affiliated with or endorsed by Project Ben-Yehuda.

Licensing

The source texts are provided according to the terms and status specified by Project Ben-Yehuda. Users are responsible for verifying that their intended use complies with the original source terms.

The processing code used to build this dataset is maintained in the GitHub repository linked above.

Citation

If you use this dataset, please cite both Project Ben-Yehuda as the original source and this dataset package:

@dataset{project_ben_yehuda_corpus_nhlocal_2026_03,
  author       = {NHLOCAL and Project Ben-Yehuda},
  title        = {Project Ben-Yehuda Corpus},
  version      = {pby-2026.03},
  year         = {2026},
  publisher    = {Hugging Face},
  dataset      = {NHLOCAL/project-ben-yehuda},
  url          = {https://huggingface.co/datasets/NHLOCAL/project-ben-yehuda}
}