PLEASE: Good infra structure base, but no over-engineering. Only what we see, that we definitely need as a base - e. g. to cover current GPT and DeepL use cases together.

The scenario we have with the linked issue TRANSLATE-3498 will in the future happen for many other MT language resources: That we use TM and/or TermCollections in translate5 as automaticaly provided training data for the MT resources.

Therefore for TermCollections the triggers pointed out in TRANSLATE-3498 should be implemented in a generic way, so that it is easy to reuse them for other MT resources like GPT.

The way the administration of MT resources connected to TermCollections or TMs should work in the same way in GUI and back–end.

Also the trigger points of changes are the same for TermCollections.

Yet, the way of training might be different in different MT resources:

• For DeepL it is necessary to delete and recreate a glossary with each change
• For GPT simply newly created terminology should be added/GPT should be informed about deletion of deleted or changed terms
• For GPT TM training should work analogous to terminology training (see above bullet)

Trigger points for TMs to update an MT resource

Same as for TermCollections except those related to editing terms in TermPortal.

In addition:

• Saving/updating a segment in translate5 editor
• Creating/updating/deleting source and/or target of translation units in TM maintenance