We're excited to introduce a new feature that allows more control over your QA workflows!
TAUS EPIC combines Quality Estimation to estimate the content quality, and Automated Post-Editing to improve it in real time.
Every segment first gets a QE score to assess its expected quality level. An acceptance threshold is defined by the user based on their requirements, and the content falling below this threshold goes through APE to get automatically improved.
Then, the APE’d content gets sent back to QE for a new quality score, and it is returned (and charged) to the client only if the new QE score is equal or better than the initial one, and the content has changed.
For example:
Up to now, APE relied on a single threshold to determine if an APE suggestion should replace the original translation. So in the above example, the APE suggestion would be discarded.
The APE suggestion in this example, though, still shows an improvement versus the original QE score. And for some use cases it can still be beneficial to use it rather than going back to the original content with a lower score.
With the new optional lower threshold feature, the user can set a different ‘lower’ acceptance threshold for APE, if needed.
Now imagine the user sets the lower APE threshold to 0.85. They will get the APE suggestions that meet or exceed this value, also when they are below the initial QE threshold.
This optional feature empowers users to fine-tune their content processing workflows with even more granularity.
Here is how this would look like in the API settings:
To improve the visibility and trackability of APE behaviour we have also added the "apeTriggered" (true/false) flag to the API response.
For more detailed information, please refer to the "Automated Post Editing" section in our API documentation.