NHLOCAL/otzar-hatorah

Otzar HaTorah Corpus

Dataset Card for Otzar HaTorah

Dataset Summary

The Otzar HaTorah Corpus is a large, comprehensive collection of contemporary Rabbinic and Torah scholarship, sourced from the "גמ"ח אוצר התורה החדש" (Gmach Otzar HaTorah HaChadash) digital library. The dataset comprises approximately 12,000 documents, originally in formats like .doc, .docx, and .pdf, which have been converted to Markdown format.

The content covers a vast range of Judaic studies, including:

• Talmud (ש"ס): Commentaries and study guides on various tractates (e.g., זבחים).
• Halakha (שו"ע): In-depth analysis of Jewish law, primarily from the Shulchan Aruch.
• General Topics (כללי): Essays, lectures, and source sheets on a wide array of subjects, from foundational Jewish philosophy (e.g., "אבות לפני מתן תורה") to practical modern halakhic questions.
• Works by Specific Rabbis and Lecturers: The collection is organized into folders representing different authors and yeshivas, such as שיעורי הרב אריאב עוזר and חבורות from הרב אריה גרבוז.

The primary language of the corpus is Hebrew. The dataset is intended for natural language processing research, digital humanities, and to serve as a valuable resource for Torah scholars.

Supported Tasks and Leaderboards

This is a raw text corpus and does not have any active tasks or leaderboards. It can be used for a variety of downstream NLP tasks, including:

• Language Modeling: Training large language models on Rabbinic Hebrew.
• Information Retrieval: Building search engines for Torah sources.
• Text Generation: Fine-tuning models to generate text in the style of Torah scholarship.
• Topic Modeling: Discovering themes and connections across the corpus.
• Named Entity Recognition: Identifying sources, Rabbis, and concepts.

Languages

The text in the dataset is in Hebrew (he).

Dataset Structure

The dataset consists of a single split, train. Each row represents one document from the original collection.

Data Instances

A typical data instance looks like this:

{
  "text": "# בס\"ד\n\n## אבות לפני מתן תורה\n\nתוכן המאמר כאן בפורמט Markdown...\n\n...",
  "source": "כללי/דור המלקטים עניינים כלליים/כללי/אבות לפני מתן תורה.doc",
  "metadata": {
    "title": "אבות לפני מתן תורה",
    "categories": [
      "כללי",
      "דור המלקטים עניינים כלליים",
      "כללי"
    ],
    "author": "otzar1",
    "created": "2005-10-31T03:44:00+00:00",
    "modified": "2018-06-04T22:52:00+00:00",
    "last_modified_by": "Klal Yisrael",
    "title_meta": "בס\"ד"
  }
}

Data Fields

• text (string): The full content of the source document, converted to Markdown format.
• source (string): The relative file path of the original document within the "Gmach Otzar HaTorah" archive. This preserves the original hierarchical structure.
• metadata (dict): A dictionary containing metadata about the source file.
  • title (string): The cleaned-up filename, serving as the document title.
  • categories (list of strings): The directory path of the file, representing its categorization within the library.
  • author (string): The author associated with the file as per the file properties. This may be a general placeholder. The true author is often indicated in the categories.
  • created (timestamp): The creation timestamp of the original file.
  • modified (timestamp): The last modification timestamp of the original file.
  • last_modified_by (string): The user who last saved the original file.
  • title_meta (string, optional): Additional metadata extracted from the document's title property, such as "בס"ד".

Data Splits

The dataset contains a single split:

Split	Number of records
train	~12,000

Dataset Creation

Curation Rationale

This dataset was curated to preserve and make accessible a vast body of modern Torah scholarship in a machine-readable format. The goal is to enable computational analysis and the development of NLP tools for the domain of Rabbinic literature, which is often underrepresented in digital corpora.

Source Data

Initial Data Collection and Normalization

The source data is a digital archive named "גמח אוצר התורה החדש טבת תשפה" (Gmach Otzar HaTorah HaChadash Tevet 5785). It consists of files created by numerous individuals and institutions over many years. The original files are primarily in Microsoft Word (.doc, .docx) and PDF (.pdf) formats.

The data was processed by converting the contents of each file into a unified Markdown (.md) format. The folder structure was parsed to generate the categories in the metadata, and file properties were extracted to populate the other metadata fields.

Who are the source language producers?

The producers are rabbis, yeshiva students, scholars, and kollel members from primarily the Haredi (Yeshivish) community in Israel and abroad. The texts represent a specific style of analytical and source-based learning known as "lomdus".

Annotations

The dataset does not contain manual annotations. The metadata fields are automatically extracted from the file system and document properties and serve as a form of structural annotation.

Personal and Sensitive Information

The documents are scholarly religious texts and are not expected to contain personal or sensitive information. However, as the collection is vast and automatically processed, it has not been manually reviewed for PII. Users should exercise their own judgment. Author names and affiliations are considered public information in this context.

Considerations for Using the Data

Social Impact of Dataset

The dataset can have a significant positive impact by making Torah scholarship more accessible for research and study. It can power new tools for students and scholars, and enable large-scale analysis of contemporary Halakhic and Talmudic thought.

Discussion of Biases

The corpus reflects the perspective and academic style of the specific religious communities (primarily Lithuanian-style Haredi Yeshivas) that produced it. It may not represent the full spectrum of Jewish thought or halakhic approaches. The terminology, choice of sources, and methods of analysis are characteristic of this tradition.

Other Known Limitations

• Conversion Errors: The automatic conversion from .doc/.pdf to Markdown may introduce formatting artifacts, OCR errors (for image-based PDFs), or loss of complex layouts (e.g., tables, marginalia).
• Inconsistent Formatting: The source documents lack a unified style, leading to inconsistencies in how sources are cited, how headers are used, and general document structure.
• Metadata Accuracy: The file metadata (like author and created date) may not always be accurate, as files were often copied and shared. The categorical data derived from the folder structure is generally more reliable.

Additional Information

Dataset Curators

The data was collected and organized by the volunteers of the "Gmach Otzar HaTorah". The conversion and structuring for this Hugging Face dataset were performed based on that collection.

Licensing Information

The dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.

Citation Information

If you use this dataset in your work, please cite it as follows:

@dataset{otzar_hatorah_corpus_2024,
  author       = {NH Local},
  title        = {Otzar HaTorah Corpus},
  year         = {2025},
  publisher    = {Hugging Face},
  version      = {1.0.0},
  url          = {https://huggingface.co/datasets/nhlocal/otzar-hatorah}
}