Practical machine learning in linguistic study: an example of deep neural network

This talk will introduce some current developments in machine learning technologies and discuss how they can help linguistic study. Machine learning methods can be roughly categorized as 1) supervised learning and 2) unsupervised learning. While supervised learning learns an abstract relationship between input and output feature mappings and requires (human-)labeled data, unsupervised learning discovers hidden patterns in unlabeled training data. I will also briefly explain the theories, limitations and possible applications of deep neural network (Hinton et al., 2006), a very powerful machine learning method, and demonstrate an example of deep neural network modeling the articulatory targets of English vowels.