Sharing is caring!

Back in 2006 when Google released Google Translate, the algorithm was running on a Phrase-Based Machine Translation. The translations offered by the service have been greatly improved in the recent years due to extraordinary progress when it came to artificial intelligence. To add to that, there have also been some additions regarding neural networks which further boosted the performance in Google Translate.

Naturally, language is phrase-based, which is why translation is a difficult area in itself. Computers have been designed to manage phrase-based translation, but there are still nuances in language that machines cannot grasp, as we’re sure many people using the translating tool have seen along the years.

But now Google made a step forward and deployed an innovative Google Neural Machine Translation system, which uses neural networks and machine learning strategies in order to boost accuracy in translations. GNMY manages to do this by including a sentence to be translated both as a whole, but also while taking in the phrases within it, in opposition to the tendency to treat words and phrases as individual components.

Neural networks find their source in the neuron structures found inside the human brain. They are based on the same AI (artificial intelligence) technology that runs the systems that identify spoken commands, places or people in an image.

The specific neural network used for Google Translate is called LSTM, which stands for long short-term memory. This type of network can hold information both on the short run and on the long one, which is a behavior similar to the human brain and memory. While the system analyzes a sentence, it can remember all the parts in it, considering it a whole instead of splitting it up in words and phrases. According to Google’s predictions, this step forward will reduce errors by 60% in some languages.