Datasets:

riotu-lab
/

MURAD

license: cc-by-nc-nd-4.0
task_categories:
  - summarization
  - text-generation
  - question-answering
language:
  - ar
tags:
  - nlp
  - llm
  - RD
  - reverse_dictionary
pretty_name: MURAD
size_categories:
  - 10K<n<100K

A Large-Scale Multi-Domain Unified Reverse Arabic Dictionary Dataset

Authors: Serry Sibaee, Yasser Alhabashi, Nadia Sibai, Yara Farouk, Adel Ammar, Sawsan AlHalawani, Wadii Boulila

Overview

MURAD A Large-Scale Multi-Domain Unified Reverse Arabic Dictionary Dataset is an open Arabic lexical dataset designed to support research in computational linguistics, lexicography, and Arabic natural language processing (NLP). The dataset contains 96,243 word–definition pairs spanning multiple scientific, religious, and linguistic domains.

Arabic is a linguistically and culturally rich language with a vast vocabulary; however, large-scale, structured lexical resources that link Arabic words to precise definitions remain limited. MURAD addresses this gap by aggregating and standardizing definitions from trusted reference works and educational sources.

The dataset is particularly suitable for:

Reverse dictionary modeling
Semantic retrieval and search
Lexical semantics research
Educational and digital humanities tools
Arabic NLP benchmarking and evaluation

Dataset Description

Each record in MURAD consists of:

Target Arabic term
Standardized Arabic definition
Source metadata, including domain, reference ID, source title, URL, and extraction method

The data were extracted using a hybrid processing pipeline that integrates:

Direct text parsing (scraping)
Optical Character Recognition (OCR)
Automated reconstruction and normalization via GPT-4o

This pipeline was designed to ensure high accuracy, clarity, and consistency across heterogeneous source materials.

Domain Coverage

MURAD covers a wide range of domains, including but not limited to:

Linguistics
Islamic studies and jurisprudence
Psychology
Mathematics
Physics
Chemistry
Engineering (mechanical, electrical, electronic)
Machine learning, deep learning, and data science
Measurement and scientific terminology

Data Format

The dataset is released in a structured, machine-readable format (CSV, UTF-8) suitable for direct use in NLP pipelines. Each entry aligns a single Arabic term with its definition and associated source metadata.

Column name	Description
`Ref ID`	Numeric identifier (1–17) corresponding to the source reference work
`word`	The target Arabic word or term
`definition`	The formal Arabic definition associated with the word
`Source Title`	Original Arabic title of the reference work
`Source URL`	Direct hyperlink to the source (online book, PDF, or institutional page)
`English Translation of the Title`	English translation of the source title
`Extraction Type`	Method used to obtain the definition, such as Scraped, Extracted, or OCR
`Domain`	Subject domain of the source reference work
`Editor`	Editor or editorial body responsible for the reference work
`Year`	Publication year of the reference work
`Authors`	Author(s) of the reference work
`ISBN`	International Standard Book Number (ISBN) of the reference work, where available

Sources and References

The following table lists the primary Arabic reference sources used to compile the MURAD dataset. Counts indicate the number of word–definition pairs extracted from each source. Reference IDs correspond to identifiers used in the dataset metadata.

Ref	English Translation of the Title	Extraction Type	Count
1	Al-Kafawi's Dictionary of Universals	Scraped	14,476
2	Al-Jurjani's Book of Definitions	Scraped	1,399
3	Dictionary of Chemistry Terms	Extracted	4,468
4	Dictionary of Machine Learning Terms	Extracted	1,758
5	Dictionary of Mathematical Terms	Extracted	7,808
6	Dictionary of Physics Terms	Extracted	5,081
7	Dictionary of Arabic Measurement Terms	Extracted	550
8	Dictionary of Psychology Terms	Extracted	4,157
9	Dictionary of Mechanical Engineering Terms	Extracted	1,569
10	Book of Terminology in Arabic Sciences	OCR	13,181
11	Encyclopedic Dictionary of Applied Linguistics Terms	Extracted	13,350
12	Dictionary of Islamic Jurisprudence Terms	OCR	9,964
13	Dictionary of Electrical, Electronic, and Communication Engineering Terms	Extracted	1,422
14	Dictionary of Scholars' Terminology	Scraped	7,907
15	Encyclopedia of Faith Terminology	OCR	4,160
16	General Terminology Dictionary	Scraped	3,750
17	SDAIA Data and Artificial Intelligence Glossary	Extracted	1,243

Intended Use

MURAD is intended for research and educational purposes, including:

Training and evaluating Arabic NLP models
Studying Arabic lexical semantics and definition modeling
Building reverse dictionaries and semantic search systems
Supporting digital lexicography and curriculum development

License and Availability

This dataset is released openly under the CC BY 4.0 license to promote reproducible research and broader access to high-quality Arabic lexical resources.
The dataset is publicly available at: https://huggingface.co/datasets/riotu-lab/MURAD

Citation

If you use MURAD in your work, please cite the following paper:

APA:

Sibaee, S., Alhabashi, Y., Sibai, N., Farouk, Y., Ammar, A., AlHalawani, S., & Boulila, W. (2026). MURAD: A Large-Scale Multi-Domain Unified Reverse Arabic Dictionary Dataset. arXiv preprint arXiv:2601.21512.
https://arxiv.org/abs/2601.21512

BibTeX:

@misc{sibaee2026murad,
      title={MURAD: A Large-Scale Multi-Domain Unified Reverse Arabic Dictionary Dataset}, 
      author={Serry Sibaee and Yasser Alhabashi and Nadia Sibai and Yara Farouk and Adel Ammar and Sawsan AlHalawani and Wadii Boulila},
      year={2026},
      eprint={2601.21512},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2601.21512}, 
}