Some time ago, a neural network based on artificial intelligence was connected to the Google Translate system to speed up translation processes into several of the most common languages. But after a while, the machine translator learned to recognize those languages that were not originally embedded in it. This became possible due to the fact that the artificial intelligence independently created an "intermediate" language, which is used to translate from one dialect to another.
Google Translate, unlike other online translators, translates the entire sentence without breaking it down into individual words. Of course, the quality is still far from ideal, but it is constantly improving. Due to the grammatical difference between languages for translating phrases and sentences, it is necessary to use different language modules, which are different for each language. The neural network, having systematized the information of these modules, has created its own "middle" version, which is used for translation.
Thus, at first the system translates the language into this “average” fictional one, and already from it it is capable of translating into any other. For example, initially the system was trained to translate between English and Japanese, as well as English and Korean. Then the system itself learned how to translate from Japanese into Korean, bypassing English. Many experts see great potential in this system. For example, Kyunghun Cho, a semantics scholar at New York University, states:
“The translation method implemented by Google's specialists is more complex than any other method. Nevertheless, this direction is developing very rapidly, and in the near future such systems will become the main means of automatic translation. I have no doubt that we will create and train an automatic translation system based on a single neural network that will master more than a hundred different languages simultaneously."
VLADIMIR KUZNETSOV