Problem

Quality estimation should be a help for post-editors and reviewers. They should see the estimation for a segment and the average for the whole task. And they should get immidiate feedback after saving a segment, how the quality estimation changes.

Solution

Probably it makes sense to use as base technology the existing framework for match analysis data storage and also display (as in a further issue we will get a quality estimation analysis tab similar to the current TM analysis)

Introduce a new column "Quality estimation".

Located by default at the right side of the "target" column.

It contains the estimation value for the segment (0 to 100). Its background color works like currently runtimeOptions.plugins.VisualReview.matchRateColorization works:

• it can be configured, what match range should have what color
• match ranges can be defined as the admin wants regarding number of ranges and where ranges start and end

runtimeOptions.plugins.VisualReview.matchRateColorization is  removed or better converted into a runtimeOptions.qualityEstimation.rateColorization (overwriteable on client, import and task-level):

• existing values are taken over
• if estimated values for segments exist, they are colored also in the visual
• In difference current system, range for green (best matches) also needs to be defined in config and nothing is shown without color
• the background of the visual is also colored, like it is the case currently
• yet the background of the visual has a stronger opacity then currently. Background color of the column in the segment table should be strong and clear and in the visual there is a strong opacity of the same colors to not interfer to much with the visual (much more opacity then currently).
• The green of the currently active segment in the visual needs to have a strong contrast to the green of the quality estimation in the visual.

Saving of segments - new calculation of quality estimation

When a segment is saved, it is immidiately send to the quality estimation resource to re-calculate the estimation for the changed target text.

The value and color in the front-end is updated for the segment as soon as it is calculated. It is communicated back to the front-end indpendent from the back-communication of the target segment itself (with the new term-markup in it ect.) to not delay the back-communication of the target segment.

A config (overwriteable on client, import and task-level) dertermines, if the new calculation actually is triggered or not.

Another config (overwriteable on client, import and task-level) dertermines, how many segments before and and how many after the current segment are passed with the current segment as context. The prompt explains the LLM, that the calculation should be done for the current segment and the rest is context.

Average estimation for the complete task

Next to the progress bar above the segment table a field will be displayed, that shows the average quality estimation for all segments. It will have a background color based on the schema described above for the segments.