“Yandex Enhances Translator Quality Using YandexGPT”

Yandex, the leading Russian IT company, has announced that they have upgraded the performance of their machine translation service, Yandex Translator. The improvement was made possible through the further training of the underlying neural network using Yandex’s large-scale language model, YandexGPT. The upgraded translator now demonstrates an enhanced understanding of context, idioms, and technical terminology.

The additional training has enabled the service to establish more accurate connections between sentences, and it now handles long texts more effectively. Thanks to its superior context analysis ability, the machine translator detects the professional focus of a text and leaves necessary terms untranslated.

Yandex has been using AI in its Translator service since 2018, training it on pairs of texts — original and its translation into another language. This is a lightweight neural network capable of processing a large number of queries in real time. In contrast, YandexGPT generates complex texts in a lexical and stylistic variety, including specific terminology. Therefore, the company decided to use the larger language model for additional translator neural network training.

Yandex engineers prepared a special version of YandexGPT, optimized for translating texts. During the pre-training stage, it processed a large volume of texts in both Russian and English, learning lexicon, morphology, and syntax. This was followed by fine-tuning of the large language model to cater to the tasks of translation. During the Reinforcement Learning (RL) stage, the translation quality of YandexGPT was assessed, and the results were ranked from best to worst.

Upon completion of the neural network’s additional training, its performance was subjected to a comparative evaluation. Experts fluent in both Russian and English compared pairs of long and complex texts translated by the two versions of the neural network. The new version outperformed the old one in 57% of cases. The upgraded platform has already been integrated into the Yandex Translator, search engine, and video translation service in the Yandex Browser.