The MT model improves daily through automated, targeted reviews that happen in the background while still delivering instant translations via the API, without sacrificing speed or scale.
Translated, a provider of AI-powered language solutions, unveiled the first dynamic, adaptive machine translation (MT) system that enables enterprises to continually improve MT quality in real-time through a unique symbiosis between humans and machines.
ModernMT’s new Human-in-the-Loop (HITL) plan allows users to strategically allocate resources and scale the involvement of translators to ensure continuous improvement in MT at a faster rate than ever before.
The MT model improves daily through automated, targeted reviews that happen in the background while still delivering instant translations via the API, without sacrificing speed or scale.
Previously, MT buyers have had to rely on periodic evaluations or direct user feedback to understand where MT translations could be improved. Even with that information, updating the model is cumbersome, if even possible. The HITL plan addresses these obstacles with a ‘Machine-First, Human-Optimised’ approach.
While suitable for all MT users, the new solution will be particularly effective for organizations that rely on raw MT output (e.g., in real-time and/or high-volume scenarios) and have no means of identifying potential issues in the output, or addressing them, in any regular or consistent way. This solution simplifies the process of monitoring quality and deciding where best to invest in improvements across languages and content types.
“Our Human-in-the-Loop plan reimagines adaptive machine translation for enterprises. It’s designed to be dynamic, flexible, and deeply integrated with translator expertise, especially for large MT-driven projects,” said Marco Trombetti, CEO of Translated. “This elegant solution enables organizations to monitor, plan, and invest in continuous MT quality improvement in a highly automated, cost-effective, and structured way.”
Translated’s HITL approach is a true innovation that combines the best elements of AI, highly skilled people, and efficient workflows, to allow enterprises to use MT to scale operations internationally.
Users can easily adjust what proportion of their machine translated content should be reviewed, from 0% (no review) upwards, in a new, easy-to-use interface. This can be done for any language combination and the user gets an instant estimate of cost variations. The system automatically selects which translations to review using a new AI-powered quality assessment feature that ultimately adapts to the users’ content to identify the most appropriate segments for review.
Reviews are carried out by professional translators selected from Translated’s community of over 300,000 vetted, native-speaking professionals. The most suitable translators for a given task are chosen by a proprietary, AI-powered tool that considers over 30 parameters, including past performance and availability. The MT engine is subsequently re-trained continuously with corrections it receives.
