How Yandex Translate makes over 7,000 publications available in many languages

Using the PressReader platform, users can subscribe to digital versions of popular newspapers and magazines on their iOS, Android, computers, and other devices.

PressReader offers over 7,000 publications from over 120 countries available in more than 60 languages, and since 2017 have been using Yandex Translate to help deliver a superior reader experience.

PressReader processes thousands of pages each and every day, extracting text and images from them. Many newspapers, including The Washington Post, USA Today, International Herald Tribune, Daily Mail, and The Herald, appear on the platform as soon as they hit newsstands. All materials become available for search, and they are easy to read on mobile devices.

PressReader makes publications available to the broadest range of readers possible. A German article published in the morning must be available within a few hours to an Indian student or an American businessman in their native languages.

No agency can translate such volumes with any real speed — what’s needed is a high-quality machine translation.

This became all the more important during the pandemic when the need for access to content online became even greater.

Price, quality, scaling, and data processing speedPrice, quality, scaling, and data processing speed

To create your own machine translation service, you would have to hire a team of developers, set up the infrastructure, and spend a lot of time on product development. The project costs alone would make it unprofitable for virtually any company.

Adopting cloud technologies can significantly reduce these costs. This also makes it much easier to store, organize, and transfer large amounts of information.

PressReader considered three options at the selection stage: Yandex Translate, Microsoft Translator, and Google Translate. They chose Yandex Translate because its quality and scalability are as good as its competitors while costing much less.

Yandex Translate is connected to PressReader via the API. Using cloud technologies, the service receives texts from the aggregator, determines the source language, and then returns a translated version.

Integrating the machine translation service took PressReader less than a week. Now, you don’t need to speak Italian or French to read the most recent editorial in La Repubblica or the latest issue of Vogue France. Despite the huge amount of work involved, Yandex Cloud processes the information at impressive speeds and displays the final results almost immediately.

Yandex Translate uses a hybrid machine translation model that includes neural network and statistical approaches. The result obtained from both systems is evaluated by an algorithm based on the CatBoost training method. The evaluation considers dozens of factors: from the length of the sentence (the statistical model is better at translating short phrases and rare words) to the syntax. Then the algorithm evaluates both translations based on a range of factors and shows the user the best version.

By using cloud technologies, PressReader allows its users to read any issue of almost any popular newspaper or magazine worldwide in over 60 languages.

The aggregator provides translations for 365 language pairs (and this figure keeps rising). Translating, for example, from Russian to French or from Finnish to Czech is no problem. Yandex Translate translates over two billion characters for PressReader every month.