Deep Learning-Based Context-Sensitive Typo Detection, Alteration in Clinical Notes

Spelling mistakes are a key issue in medical analysis of natural language, due to the high ubiquity of misspelled words in health records. In spite of the fact that many Natural Language Processing (NLP) systems are vulnerable to adversarial examples because they were trained on grammatically sound text data, processing correctly spelled text is essential for learning. In this research study, we propose the Conditional Autonomous Framework (CAF), a probability-based template for correcting misspelled words that evaluates the posterior distribution of the appropriate word given the misplaced apostrophe as well as the perspective. Our experiments demonstrated that our approach can effectively correct the errors.