A Benchmark Corpus of Yemeni Proverbs for Figurative and Cultural Language Modeling

We present a structured corpus of 5,252 Yemeni Arabic proverbs paired with explanations in Modern Standard Arabic (MSA). The dataset was compiled from four printed proverb anthologies and three publicly accessible online repositories between January and June 2024. All entries were manually transcribed or programmatically extracted and subsequently verified to ensure accuracy and fidelity to the original sources. Each record includes the proverb text, its explanation, source attribution, and available geographic metadata. The corpus addresses the scarcity of dialect-specific Arabic resources and provides a benchmark for evaluating figurative language understanding and culturally grounded language generation. Technical validation through supervised fine-tuning and unsupervised clustering demonstrates the dataset's internal consistency and thematic diversity. The dataset is publicly available via Zenodo under a Creative Commons Attribution 4.0 International (CC BY 4.0) license.