Probably will be/should be solved with linked analogous issue for DeepL

Problem

Terminology that is unwanted in source but preferred/allowed in target is not send to OpenAI

Example:

source terms DE are

Produktservice - forbidden

Produkt-Service - preferred

Produkteservice - admitted

target terms EN are

product service - preferred

products service - allowed

In the source text all German versions could be used and information is lost about terminology if e.g. the forbidden term was used.

Solution

If any source term is detected in the source text, provide the preferred/admitted (best category) target term for the model.

Change logic in 
MittagQI\Translate5\Plugins\TermTagger\TerminologyProvider