GoEmotions-RU: A Rehydratable Roman Urdu Emotion Dataset (ON/OFF Ablation)

GoEmotions Roman Urdu (Rehydratable) — Ablation ON/OFF is a Roman Urdu emotion classification dataset derived from the HuggingFace GoEmotions benchmark (simplified configuration). The release provides two ablation views: OFF, containing a single primary Roman Urdu translation per example, and ON, which expands each example with multiple Roman Urdu spelling variants to support robustness and ablation studies. To comply with the licensing terms of the original dataset, this Zenodo release does not include the original English text. Instead, it follows a rehydratable release strategy: each record is identified by a stable global_id that allows users to locally download GoEmotions from HuggingFace and reconstruct aligned English–Roman Urdu pairs using the provided rehydration script. The dataset preserves the original train/validation/test splits, includes normalized Roman Urdu text and emotion labels, and is intended for academic research in emotion analysis, low-resource NLP, and cross-lingual transfer. A quality-assurance report, checksum file, and reproducibility scripts are included to support transparent and verifiable use.