What is the problem you’re addressing?

“Typically, MT users have to rely on periodic evaluations or direct user feedback to understand which areas needed improvement. Even with that information, updating the model is cumbersome, if possible at all. The Human-in-the-Loop solution addresses these issues with a ‘Machine-First, Human-Optimized’ approach. While suitable for all MT users, this new plan is particularly effective for enterprises that rely on raw MT output, for example, in real-time and/or high-volume scenarios, and have no means of identifying or addressing potential problems in the output on a regular or consistent basis. It simplifies the process of monitoring quality and deciding where to invest in improvements across languages and content types.”

How does this approach benefit customers?

“Better quality, faster! This is the first dynamic and adaptive machine translation solution that enables enterprises to improve MT quality on a continuous (daily) basis, in the background, without any intervention needed. We do this through a unique symbiosis between humans and machines, allowing users to strategically allocate resources and scale translators’ involvement to ensure continuous MT improvement at a faster rate than ever before. Essentially, the MT model now improves continuously through automated, targeted reviews of low confidence translations that take place in the background on an ongoing basis. This allows for instant translations via the API without sacrificing speed or scale, supercharging your ROI in MT.”

It sounds complicated. How does it work?

“It’s actually very straightforward. The process is managed through a simple new interface in ModernMT. Users can set the percentage of their machine-translated content to be reviewed and revised for the purpose of improving the model. This can be done for any language combination, and the user will receive an instant estimate of cost variations. After that, you use the MT as normal and everything else happens automatically behind the scenes. The solution uses an new AI-powered quality estimation step to automatically select which translations require review, ultimately adapting to the user’s content to identify the most appropriate segments. Reviews are carried out exclusively within the Translated environment, by professional translators selected from Translated’s community of over 300,000 vetted, native-speaking professionals. The MT engine then continuously adapts to the corrections it receives.”