YandexGPT has taken the leading position in the global ranking of English-to-Russian translation, according to a DiBiMT study. The implementation of this large language model in Yandex Translate has doubled the accuracy of translations and made them more natural, with improved consideration of stylistic features and set expressions.
In the study, which involved 22 translation systems, the new version of Yandex Translate outperformed well-known services such as Google Translate, ChatGPT, and DeepL. Previously, Yandex used neural machine translation, which analyzed context only at the level of individual sentences. Thanks to YandexGPT, which can process the entire text, it was possible to achieve a more accurate reproduction of polysemantic words and improve the coherence between sentences.
Test results showed that the number of errors was halved, and the smoothness of the translation increased by 15%. The integration of YandexGPT already allows users to translate texts using a smart camera and study summaries of pages through Yandex Search. In the future, the model will be implemented in other company services as well.
Testing was conducted on several thousand sentences with homonyms, which allowed for an objective assessment of translation effectiveness in complex contexts. Yandex Translate with YandexGPT demonstrated better results even compared to powerful systems such as GPT-4 and Llama-3 8b, which confirms its leadership in the field of machine translation.
Read more on the topic:
Yandex Launches YandexGPT 4: A New Generation of Language Models
Yandex Presents Improved Photo Translation for Working with Texts in Images
