RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2

Yandex.Translator

Product
Developers: Yandex
Last Release Date: 2024/06/07
Branches: Internet services
Technology: Speech technology

Content

2024: YandexGPT Integration

On June 7, 2024, Yandex presented an updated version of machine translation, trained using YandexGPT. The large language model YandexGPT has prepared reference examples of texts on which the neural network in Yandex Translator was trained. This allowed the service to more accurately understand the context, recognize phraseologisms and navigate professional vocabulary.

As reported, the translator has become better at dealing with long and complex texts. Through further training, he defines the relationships within and between sentences. In addition, the quality of translation of narrowly focused articles has increased. For example, the service will now understand from the context that we are talking about a programming language or stock index, and will leave their names without translation.

source = Yandex
Example of Translation Improvement

Yandex has been using neural network-based machine translation technology since 2018. The neural network in the Translator was trained only on pairs of texts - originals and their translation into another language. So she learned to see the relationships between them and imitate them. The neural network working in the service is lightweight: this allows her to cope with a large number of user requests in real time.

source = Yandex
Example of Translation Improvement

The large YandexGPT language model is capable of generating complex, lexically and stylistically diverse texts, including with specific terminology. Therefore, the company used its potential to create reference examples. So the neural network in the Translator did not lose the speed of its work, but became smarter.

source = Yandex
Example of Translation Improvement

To generate reference training examples, Yandex has developed a specialized model of the YandexGPT family, adapting it to translation tasks. First, the model during the pretraining stage analyzed many texts in English and Russian and studied the rules of vocabulary, morphology and syntax. The next step was to fine-tune the language model (supervised finetuning) for translation tasks. Then, at the reinforcement learning stage, AI trainers assessed the quality of YandexGPT translation and ranked its responses from the best to the worst.

The work of the completed neural network in the Translator was assessed by the Side by Side (SbS) method. Assessors, who speak Russian and English, compared pairs of long and complex texts translated using two versions of the technology, and chose the best option. In 57% of cases, this version coped better.

source = Yandex
Example of Translation Improvement

The updated technology is used to translate texts from English into Russian in Yandex Translator, Search, as well as in video translation in the Browser.

2021: Yandex showed the technology of voice-over translation of video from English

In mid-July 2021, Yandex introduced voiceover technology for videos in foreign languages, which, according to the company, has no analogues in the world. The system combines machine translation, biometrics, speech recognition and synthesis technologies.

Biometric solutions allow you to determine the gender of the speaker and choose the appropriate voice. And in order for the voiceover to coincide with the picture, the translator pauses, slows down somewhere, and accelerates speech somewhere. The fact is that in the original and translation, the replicas often do not coincide in length, so a special algorithm monitors the pace of speech, Yandex explained.

Yandex demonstrated the technology of voice-over translation of video from English

Using the technology, you can watch video recordings with voice-over translation. By July 16, 2021, she works only with videos in English. The developers have already tested it on videos on various topics. This is a video about climate change, about machine learning, the history of Pluto and other things. Users of the desktop Yandex.Browser can watch the video with voice-over translation here.

File:Aquote1.png
There is a lot of useful content on the Internet that is not available to people due to the language barrier, "says David Talbot, head of natural language processing at Yandex. - And we are close to finally erasing all the boundaries. Yandex.Browser has long been able to translate texts, this year it began to translate images, video translation is the next stage. This is a big difficult task that no one in the world has yet solved.
File:Aquote2.png

As reported in Yandex on July 16, 2021, soon users will be able to choose for themselves which videos to translate using the new technology. It is noted that several development teams worked on the creation of the system.

Earlier in 2021, Yandex.Browser added a built-in translation of pictures. The function works directly on the pages of sites - the browser itself finds pictures with foreign text, and then allows you to translate them in one click.

2020

Adding the Yakut language

On April 27, 2020, it became known that Yandex included the Yakut language in the Yandex.Translator service.

Adding the Yakut language

Now any Yakut word or phrase can be automatically translated into one of the 98 languages ​ ​ of the world - or, conversely, find out how this or that expression or word will sound Yakut.

File:Aquote1.png
I think this is a very important moment in the development and preservation of the language of the Sakha people in cyberspace - first of all, it is necessary for our future generations,
noted the Head of Aysen Nikolaev the Republic in his account in Instagram
File:Aquote2.png

The initiator of support for the Yakut language in machine translators is Alexey Ivanov, a researcher at the National Library of RS (I).

File:Aquote1.png
The idea to create a machine translation for the Yakut language arose back in 2017. Over the past year and a half, bilingual Russian-Yakut texts have been collected on a voluntary basis, reference Yakut-Russian translations have been prepared to assess the quality of translation,
notes Alexey Ivanov
File:Aquote2.png

In his opinion, machine translators are one of the important tools for preserving languages.

File:Aquote1.png
The quality of machine translation for the Yakut language is already approaching human, but still requires improvement. In my opinion, Yandex is the only company in the world that is working to preserve the languages ​ ​ of the peoples of Russia using machine translation,
tells Alexey Ivanov
File:Aquote2.png

Further plans of the project team include the digitalization of the cultural heritage of the indigenous peoples of the Arctic.

The content of the RS (I) electronic library for April 2020 is more than 34,000 digitized documents, readers are given 8-9 thousand e-books a day.

The potential of the National Library of RS (I) allows you to create a Center for the digitalization of the linguistic and cultural heritage of the peoples of Yakutia on the basis of digitized text, audiovisual, archival, museum artifacts collected into a single digital platform.

AppGallery Availability

On April 8, 2020, it became known about the full integration company mobile applications Yandex in the app store. AppGallery More. here

2019: Real-time speech translation function launched

On August 25, 2019, the Yandex.Translator service introduced a real-time speech translation function. In cases where you need to discuss something with a foreigner, make an order in a restaurant, find out the way from passers-by or explain yourself with a taxi driver abroad, the application will act as an interpreter: it will listen to replicas, translate them and voice them to the interlocutor.

Yandex.Translator now translates speech in real time

To start a conversation, you need to choose a communication language for yourself and the interlocutor - and you can speak by clicking on the flags of the languages ​ ​ in turn. The application will instantly translate replicas, display them on the screen and voice them in the selected language.

If the interlocutor does not hear something, he can always be shown the text or re-voiced with one touch. If Yandex.Translator does not exactly translate the phrase, the text can be corrected. The entire dialog is saved in the story for further viewing.

In the application Android for dialogue mode, about 50 European and Asian languages ​ ​ are available, and in iOS the -version - more than 20. Among them are Russian, English, Turkish, French, Italian,, Spanish German, Chinese, Japanese and others.

The application recognizes and synthesizes speech using SpeechKit technology and smartphone capabilities. Neural networks trained on large corpus of texts are used to translate replicas.

In addition to the new dialogue mode, Yandex.Translator is equipped with the function of translating a phrase in photos, supports voice input and offline translation. The application supports 95 languages. It has a variety of selections - for example, with idioms and phrases for a resume. With its help, it is convenient to learn words and compile your own collections.

The updated application is available for download in the App Store and Google Play. If Yandex.Translator is already installed, the dialog mode will appear immediately after the update.[1]

Notes