Text Corruptor

Text corruption is surprisingly useful. While most tools focus on cleaning or formatting text, intentionally introducing errors has several practical applications in software development, machine learning, and design.

For developers building spell checkers, autocorrect systems, or search engines with fuzzy matching (like Elasticsearch or Algolia), you need test data that contains realistic human mistakes. If your search engine can't find "restaurant" when a user types "restarant", you need to know that beforehand. Generating these typos manually is tedious and rarely covers the true variety of human error.

In machine learning, "data augmentation" is a common technique used to make natural language processing (NLP) models more robust. By training models on text that contains typos, missing letters, and grammatical errors, the models learn to handle messy, real-world input without failing. It's the same principle as training self-driving cars with images of obscured stop signs.

This tool lets you introduce 8 distinct types of errors: from simple mechanical typos (swapping adjacent keys) and missing letters, to more cognitive errors like homophone substitution (replacing "their" with "there") and transposed words. You can control the intensity from a subtle 1 (occasional mistakes) to a chaotic 10 (barely readable). All processing happens instantly in your browser, generating a unique variation every time you click Regenerate.