Yandex has created a «hybrid» artificial intelligence for translation of texts

© Fotolia / taa22Человеческий brainYandex has created a «hybrid» artificial intelligence for translation of texts© Fotolia / taa22

The specialists of «Yandex» has created a new system of online transfer, which uses both old and statistical algorithms, and artificial intelligence, the press service of the company.

«While our new system only supports translation from English into Russian language. Soon we will begin to add other language pairs and in a few months implement support for the most popular languages,» — said the press service of «Yandex», answering the questions of the correspondent of RIA Novosti.

In the past few years leading IT companies started to develop and use various forms of artificial intelligence to search for information in the Network, processing photos and catalog them, and many other purposes where non-classical data processing algorithms do not work.

A striking example of this approach was the artificial intelligence system AlphaGo, openly beat the world Champions and European and anonymously winning hundreds of top players in the online system of this ancient Chinese game.

Most of these developments are based on the so-called deep or convolutional neural networks. They represent cascades or layered structure of several tens or hundreds more simple neural networks. Each of them handles not the original data and products of analysis of the obtained network, located above, which allows to simplify very complex problems and solve them using relatively modest computational resources.

Something similar happens when you use artificial intelligence to translate from one language to another. When the neural network «reads» the text to be translated, it does not attempt to translate it for phrases or individual words, and searches for the answer as a whole, using the data accumulated in the analysis of texts translated by human.

In this approach, as noted by the specialists of «Yandex», there are both advantages and disadvantages: in this form, the text becomes more readable and similar to how it would translate man, but, on the other hand, the car periodically starts to «dream», if rare words and phrases.

This issue of «Yandex» decided by combining artificial intelligence with the old transfer system, cope well with rare words, but are faced with great challenges when combining them into sentences. When the user enters text, it simultaneously translates and neural network and a statistical algorithm, then the system machine learning CatBoost, the newly established «Yandex», compares the results and makes them more readable and competent option.

Now the old version of «Yandex.Translator» supports 94 language, including a number of rare and fictional, such as Sindarin — the language of the elves from the books of John. R. R. Tolkien.