Does entities and synonymes helps in intention recognition

I do not have a precise example for this question, I was wondering if adding entities and / or synonymes would help the NLU for intention recognition. It seems to me that it should, particularly the synonymes who could prevent having 10 trainings sentences with nearly the same words and just one word changing.

A second question is, do I need the same number of training phrases for each intention ? I tried to search for this answer, but didn’t find one. Do I need a homogeneous set of training or is it not important as long as I have a sufficient number of training samples (let’s say at least 30 for each intention, with some intentions having +50)

right now, entity extraction is independent of intent classification.

it is recommended to have roughly the same number of sentences, but not necessary, if performance is satisfactory

Is this still true for Rasa 2.x. Are the synonym entities improving intent recognition or not?

Right now we added several synonym entities to out Bot and labeled examples in our training-data in the hope that this would help intent recognition. Assuming that if the user types in the same sentence just with a different synonym for a key-word, the intent would still be recognized correctly.