Problem

For MQM based TQE the calculation of the TQE score is displayed in the reasoning column whereas the reasoning of the LLM plus alternative suggestions are only available in the tooltipp for each MQM criterion. There copying is not possible and usability is poor.

When a segment is edited by a human and TQE per segment is disabled, the reasoning will be deprecated, but user takes no notice.

When TQE is part of the import process, the column segment at time of import is not reflecting this.

Solution

For each error criterion found in the segment, and in the order of the occurrence, the information from the tooltipp should be moved to the reasoning column like this:

13 Grammar, Minor

13 Use: machen, bevor

13 Why: A comma is required before the subordinate clause.

2 Mistranslation, Major

2 Use: ab September

2 Why: The source specifies September, not August.

Tooltipp for each MQM tag:

MQM: 2 issues, 6.0 penalty points / 28 words — Grammar/minor (1.0), Mistranslation/major (5.0)

See reasoning column for detailed information.

This only applies for MQM TQE, the classic LLM TQE should not be affected by these changes.

Important 

We do allow a human-in-the-loop scenario: LLM applies MQM tags, adds reasoning and suggests better alternative as described above. Next thing a human may check the MQM tags manually and adjust them, whenever they seem wrong. The reasoning text in the reasoning column may then be deprecated! Suggestion:

In the above example, human decided that 13 is not an error and deletes MQM tag completely. Add to the reasoning column, but keep the rest:

Segment changed, LLM reasoning deprecated!

13 Grammar, Minor

13 Use: machen, bevor

13 Why: A comma is required before the subordinate clause.

2 Mistranslation, Major

2 Use: ab September

2 Why: The source specifies September, not August.

We can do so, whenever a segment is changed. This could even be done when somebody post-edits a segment. No harm to label the TQE reasoning as deprecated. Even for classic TQE. Only exception: When TQE per segment is allowed, because then TQE reasoning is always up to date.

When TQE is part of the import process

When TQE has been done right after pre-translation as part of the iport process, then the initial translation + MQM tags must be reflected in the "segment at time of import/pretranslation" column. Resetting the segment to initial content must then reset everything as in the column "at time of impirt" and original reasoning (no deprecated info) must be restored.