license: mit
Spell-Check Dataset
This dataset consists of pairs of misspelled words and their corresponding correctly spelled words, designed for training and evaluating character-level spelling correction models. It is particularly useful for tasks such as:
Spelling correction
Character-level sequence-to-sequence modeling
Error detection and correction in text
Each data point in the dataset contains:
misspelled: A misspelled version of a word.
correct: The corrected spelling of the word.
Notes
The words were derived from the NLTK words corpus, which provides a comprehensive list of English words.
Synthetic Misspellings
Misspellings were programmatically generated using common spelling error patterns, including:
Character swaps: e.g., teh → the.
Extra characters: e.g., thhe → the.
Missing characters: e.g., th → the.
Keyboard proximity errors: e.g., grerb → gerb.
Citation
@dataset{etheridge2025spellcorrection,
author = {Torin Etheridge},
title = {Spell-Correction Dataset: Misspelled to Corrected Word Pairs},
year = {2025},
publisher = {Hugging Face},
journal = {Hugging Face Datasets},
url = {https://huggingface.co/datasets/torinriley/spell-correction},
note = {A dataset for training and evaluating character-level spelling correction models.}
}