Published on

What is Machine Translation?

Authors
  • avatar
    Name
    Antonio Castaldo
    Twitter

It comes often with difficulty explaining where my research field stands to non-experts, at the intercection of language and computer science. This does not come by surprise, considering we have always been taught that languege is an inherently human ability.

As language and technology continue to intertwine, we are increasingly aware of how powerful and helpful machines could be, if given the ability to use natural language. In this blog, I will share my knowledge and code so that other interested people may benefit in my passion for natural language processing and artificial intelligence.

We have come a long way since George Artstrouni, a georgian engineer, developed his first prototype of what he called a "mechanical brain" in 1932 1. Artstrouni's mechanical brain was a multilingual dictionary-based machine, capable of providing raw word-by-word translations.

Since Artstrouni's invention, we acquired the ability to translate from and into 200 languages and we are increasingly capable of translating creativity, figures of speech and figurative language. Of course, this was not before the discoveries in the field of neural networks, with incredible scientists and researchers such as Frank Rosenblatt with his Perceptron in 1958, Geoffrey Hinton with his work on backpropagation algorithms 2 and more recently, Google DeepMind with the world-changing research on Transformers 3 by Vaswani et al. (2017).

In the next blog posts, I would like to start not too far away in history. This way I will be able to soon start discussing state-of-the-art technologies in machine translation and NLP, and recent research developments in the relevant fields. Hence, although the next blog posts will introduce the readers to the developments in neural networks, we will not delve into the basics of artificial intelligence.

I will give for granted that any reader will be accustomed with knowledge to answer the following questions:

  • What is a Perceptron?
  • What is Gradient Descent?
  • What is an activation function?
  • What are weights and what do they represent?

Here are some resources to learn more about these topics:

  1. A Comprehensive Guide to the Backpropagation Algorithm in Neural Networks
  2. IBM's Introduction to Gradient Descent
  3. Quick Introduction to Weights (Gradients)

Footnotes

  1. Hutchins, J. (2002). Two precursors of machine translation : Artsrouni and Trojanskij. ↩

  2. Rumelhart, D.E., Hinton, G.E., & Williams, R.J. (1986). Learning representations by back-propagating errors. Nature, 323, 533-536. ↩

  3. Vaswani, A., Shazeer, N.M., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., & Polosukhin, I. (2017). Attention is All you Need. Neural Information Processing Systems. ↩

Discuss on Twitter • View on GitHub