How Has Deep Learning Revolutionized Human Language Technology?

• Published in: IEEE Signal Processing in Medicine and Biology Symposium pp. 1 - 2
• Main Author: Ma, Tao
• Format: Conference Proceeding
• Language: English
• Published: IEEE 05.12.2020
• Subjects: Computational modeling; Gaussian mixture model; Hidden Markov models; Linguistics; Natural language processing; Production; Signal processing
• ISSN: 2473-716X
• DOI: 10.1109/SPMB50085.2020.9353633

Communication by voice and text, which we refer to as natural language, is a skill that separates humans from all other species. Only humans possess the complete linguistic package. Since the first computer was invented, it's been our dream to interact with computers using spoken and written words. For decades, machine learning approaches in automatic speech recognition (ASR) and natural language processing (NLP) have been based on shallow models such as Gaussian Mixture Models (GMMs) and hidden Markov models (HMM). These approaches use hand-crafted features and models that attempt to integrate knowledge of speech production, perception and linguistics. We often refer to this knowledge as subject matter expertise. Integration of such knowledge has been a cornerstone of signal processing research for decades.