EPIC Batch API for Asynchronous Quality Estimation

Up until now, when using EPIC we were processing your content synchronously by default. This means you were getting our QE scoring in real time.

But for some use cases that might not be the best way to go, especially when large volumes of text need to be processed. 

With our new Batch API, we’ve added the option to submit your content for asynchronous processing. This is ideal for high-volume, non-interactive workflows where instant responses are not required.

In this guide you can find more information on the two APIs (real-time and asynchronous) and when to choose which. Here you can find the full technical documentation on the Batch API. Check out pricing for batch mode processing here.

Improved language Identification (LID) support

A feature that was highly requested is Language ID Support. While our model is language-agnostic, meaning that it can handle any language combination, you still want to be able to identify segments that are not translated.

With this new language identification support, the  Quality Estimation score for a target segment will be penalized if the API detects that the language of the target segment does not match the language code included in the QE request.

This feature is currently enabled by default, but can also be disabled depending on the content requirements.

More technical information on this is available in our documentation.

RAG support for APE 

TAUS has a variety of prompts prepared specifically to improve lower scoring QE segments via automatic post-editing. However, you might prefer to use your own termbases or glossaries to ensure that any APE’d segments fit your style. 

In this new release, we’ve introduced the possibility to upload glossaries and TMs as additional reference material to improve the APE results and make them more consistent with your content.

You can find here all information on how to use this new feature.