metadata
license: mit
task_categories:
- text-generation
language:
- en
tags:
- typo
pretty_name: M2M
size_categories:
- 10K<n<100K
Clear Spelling Dataset
Overview
The Mistake to Meaning (M2M) dataset is a carefully crafted synthetic collection of 100,000 unique English spelling mistakes and their correct forms, intended for training high-quality typo correction and spell checking AI models. It covers various types of common mistakes observed frequently in real-world scenarios, such as:
- Keyboard adjacency typos
- Letter swaps and omissions
- Duplicate characters
- Phonetic substitution errors
- Commonly confused homophones (e.g., "their" vs. "there")
Dataset Format
The dataset is provided in CSV format with two clearly defined columns:
| Column | Description | Example |
|---|---|---|
error |
The misspelled or incorrect word or phrase | "teh" |
correct |
The correct word or intended phrase | "the" |
Usage
This dataset is ideal for:
- Training and fine-tuning typo correction models
- Benchmarking spell-checking algorithms
- Enhancing NLP model robustness to real-world noisy input
Quality Assurance
- No duplicates: Each (error, correct) pair is unique.
- Hand-curated seed set: Includes hundreds of common misspellings verified against real-world usage patterns.
- Realistic noise generation: Uses realistic error transformations mimicking genuine human typing behavior.
License (MIT)
This dataset is released under the permissive MIT License, which allows commercial and non-commercial use, distribution, and modification. Attribution is required:
Citation
If you use this dataset in your research or projects, please provide attribution similar to:
This [your project type] uses the Mistake to Learning dataset by ProCreations.
Enjoy training your typo-correction models!