herwoww/MultiDiac

Are LLMs Good Text Diacritizers? An Arabic and Yorùbá Case Study Hawau Olamide Toyin, Samar Magdy, Hanan Aldarmaki We investigate the effectiveness of large language models (LLMs) for text diacritization in two typologically distinct languages: Arabic and Yoruba. To enable a rigorous evaluation, we introduce a novel multilingual dataset MultiDiac , with diverse samples that capture a range of diacritic ambiguities. We evaluate 14 LLMs varying in size, accessibility, and language… See the full description on the dataset page: https://huggingface.co/datasets/herwoww/MultiDiac.