lunahr/normalization-data-mixed

--- license: mit --- # Normalization Dataset (Mixed) This dataset is a collection of 50000 rows originating from various sources: - [Wikipedia](https://huggingface.co/datasets/wikimedia/wikipedia) - 20000 rows - [PersonaChat truecased](https://huggingface.co/datasets/bavard/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.