lunahr/

normalization-data-mixed

This dataset is a collection of 50000 rows originating from various sources:

Wikipedia - 20000 rows PersonaChat truecased - 20000 rows Synthetic edge case data - 5000 rows Synthetic quoted text data - 5000 rows

The synthetic data has been generated using GPT-5.3 models. The other data was sourced from the original Hugging Face sources. This dataset can be used to train text normalizers that convert badly formatted English into correct English.… See the full description on the dataset page: https://huggingface.co/datasets/lunahr/normalization-data-mixed.