Optimizing Neural Network Language Models for Healthcare - A Focus on Speech Recognition and Spelling Correction

This study presents a novel approach to enhancing neural network-based language models for healthcare applications, particularly focusing on medical speech recognition and spelling correction. By introducing a tokenization strategy that segments words into prefix, stem, and suffix components, the model achieves reduced computational complexity while improving key performance metrics like perplexity and word error rate. Evaluations conducted on English and Arabic datasets demonstrate substantial advancements in accuracy and efficiency, highlighting the model's suitability for real-time applications in healthcare.