Translation history that apply Artificial Intelligent (AI)

Being initiative by Google’s Neural Machine Translation (also known as Google Translate), today AI technology using artificial neural networks is owning outbrreak development with the change of times.

Let’s take a look at the history from conventional automatic translation technology to AI translation.

■ Rule Based Machine Translation

This method interprets and translates sentence syntax based on previously registed rules. The rules here are the grammar rules used for translation purposes, and a large number of such rules are needed for machine translation. There are more than 1,000 rules, not to mention the rules are crossed, so the construction is quite difficult and complicated. Now, this method has become obsolete.

■Statistical Machine Translation

This method registers a large amount of bilingual data and generates translated sentences statistically. Collecting a large amount of high-quality text will further improve accuracy.

■ Neural Machine Translation

This is a translation method using artificial neural networks and deep learning. With normal translation, sentences are broken down and translated into parts, but by using NMT, the translation uses neural network learning to convert from source language → intermediate expression → target language. The results are based on the entire source sentence, so the translation results are more natural.

Currently, Google Translate is using this method.

■ The future of neural machine translation

The accuracy of machine translation is significantly improved, thanks to the introduction of artificial neural networks. Therefore, companies in many countries around the world have begun to consider the application of machine translation. When they invest in this app, they get instant translation results. The application is even more effective as it can handle multilingual documents, and give instant translation results.

The number of computational formulas of neural machine translation is said to be about 1,000 times that of RBT and SBT. Therefore, it is necessary to equip a high-powered GPU to be able to support the application.

Machine translation is forecast to be more and more natural and accurate, as technological progress is changing day by day.