chapter one

1 Deep learning for NLP

 

This chapter covers:

  • A short roadtrip through machine learning applied to NLP.
  • A brief historical overview of deep learning.
  • An introduction to vector-based representations of language.

Language comes naturally to humans, but is traditionally hard to grasp for computers. This book addresses the application of recent and cutting-edge deep learning techniques to automated language analysis. Deep learning has emerged in the last decade as the vehicle of the latest wave in AI. Results have consistently redefined the state-of-the-art for a plethora of data analysis tasks in a variety of domains. For an increasing amount of deep learning algorithms, better-than-human (human-parity or superhuman) performance has been reported: for instance, speech recognition in noisy conditions, and medical diagnosis based on images. Current deep learning-based natural language processing (NLP) outperforms all pre-existing approaches with a large margin. What exactly makes deep learning so adequate for these intricate analysis tasks, in particular language processing? This chapter presents some of the background necessary for answering this question. We will guide you through a selection of important topics in machine learning for NLP.

Figure 1.1. Chapter organization.
mental model chapter1 all revised

1.1 A selection of machine learning methods for NLP

1.1.1 The perceptron

1.1.2 Support Vector Machines

1.1.3 Memory-based learning

1.2 Deep Learning

1.3 Vector representations of language

1.3.1 Representational vectors

1.3.2 Operational vectors

1.4 Vector sanitization

1.4.1 The Hashing trick

1.4.2 Vector normalization

1.5 Summary

1.6 Further reading