Problem

On pretranslation term collections are different then other language ressources, as there can be multiple fitting target terms.

• The sub-language penalty is not respected, e.g. for en-gb task the en-us term is used for pretranslation.
• When there are multiple target terms the best one should be used

Solution

The sublanguage penalty should be respected. 

With multiple target terms the best one should be used according to the glossary creation logic for LLM/machine translation.