How to handle medium confidence segments in reviews

If you’re looking for information on blocked segments in review tasks, check out our article about ICE matches here.

Medium confidence segments are similar to ICE matches in that they have been assessed to have very high quality. However, unlike ICE matches they do not require context and can appear anywhere, with the exception of those segments which have character limitations and formatting tags. Most importantly though, the customer has not requested to block these segments like they have with ICE matches.

There are two different types of medium confidence segments in review tasks: exact matches and AI-verified segments.

Exact matches occur when exactly the same segment has been previously translated by a member of our community.

AI-verified segments are segments that our quality estimation (QE) systems have assessed automatically to be of a high quality, such that they should not require extensive editing, if at all.

Although you won’t be able to see which is which on the interface at the moment, we just wanted to give you an idea of what’s informing our systems. For more information about how these segments are assessed by AI for quality, please check out our article on quality estimation here. Remember that you will still be able to edit every segment you want to with the exception of ICE matches.

How do I approach a medium confidence segment?

Here’s an example of a task with both ICE matches and medium confidence segments:

You’ll see that the medium confidence will appear in grey, but will be editable when you click on them, at which point they will turn white while you’re working on them, like this:

It’s up to you how you prioritise your time, but we think the best way to look at these would be to assume they’re not a huge priority and focus on the ones where our systems have assessed that the translation needs a lot of work. You might wish to do this in the order on the page, or work on those segments in white before returning to the ones in grey - it’s really up to you!

What if there are mistakes in a medium confidence segment?

Well, firstly you can and should edit it! If these jobs are evaluated, these segments would be assessed along with everything else (except blocked segments).

When we highlight these segments, it’s not because we think our AI systems are infallible. We’re really trying to give you an indication of how to prioritise your work, but that’s all it is: a rough idea.

AI translation isn’t perfect - we all know that. We think, however, that learning to work with its imperfections and using the available tools to your advantage can help you and improve your experience of using the Unbabel platform. So try it out and see how you get on!