Unveiling Hyper Adaptive

State-of-the-art custom adaptive MT solution for translating billions of words at ultra-fast speeds without compromising quality.

Hyper Adaptive is ModernMT's new custom adaptive machine translation (MT) model, designed to deliver highly accurate, ultra-fast translations at scale.

The model is trained on the customer's data and configured for specific use cases. Users can train it further in real time with corrections and translation memories.

This state-of-the-art solution is for companies with specialized, unique content and high volumes of ongoing content that must be translated quickly.

Rome – October 25, 2023

Translated today announced a new model of ModernMT, its adaptive machine translation (MT) system. The new model, called Hyper Adaptive, enables companies to translate billions of words at ultra-fast speeds without compromising quality. It is domain-specific and designed for use cases such as translating user-generated content, datasets for multilingual large language models, and web content for data mining activities.

In recent years, companies have approached Translated with requests to leverage the accuracy of ModernMT's adaptive MT system to quickly translate specialized, unique content and high volumes of ongoing content. While a generic adaptive MT model can handle the request, it is not designed to translate millions of words per minute in a specific domain.

Hyper Adaptive solves this issue by using sophisticated compression techniques and training the MT model for specific use cases based on the customer's previous translations and translation memories (TMs).
The resulting MT model is much smaller than a generic adaptive model and can process content at ultra-fast speeds, in as little as 50ms for a typical sentence. Combined with Translated's dedicated data centers, it can translate the entire English Wikipedia (4.4 billion words) into another language in less than a day (3 million words per minute). By training directly using customer data, the Hyper Adaptive model achieves translation accuracy equal to or better than state-of-the-art custom adaptive MT models.
This solution helps companies maintain quality even when translating massive volumes of content at ultra-high speeds.

In some specific cases, such as user-generated content, combining the MT model with professional translators can further improve the quality of the output over time. Because the model is still adaptive, it can continue to improve after initial training through corrections and new TMs delivered to match the company's style.

We designed the Hyper Adaptive model to enable the translation of content that has never been translated before. Its language coverage allows companies to reach over 99% of the world's population in their own language. Hyper Adaptive is one more step towards global understanding.
Marco Trombetti – Translated CEO

Integration and Costs

Like all other ModernMT models, the Hyper Adaptive model can be integrated into the translation workflow via API. Costs vary depending on the use case, the amount of data to be translated, and the amount and quality of existing translations and TMs. Existing Translated customers can contact their account manager to get a new service quote.

Thanks to the Hyper Adaptive model, user-generated content on Airbnb has reached an unprecedented level of quality, greatly improving the experience for our user base. The real-time, high-quality translation of UGC has helped Airbnb foster a stronger sense of community among our hosts and guests, which has had a tremendous impact on our business.
Salvo Giammarresi – Head of Localization at Airbnb

Get started with
Hyper Adaptive

Companies interested in an ultra-fast, hyper-customized model
for massive volumes of content can contact us.

Contact us