tokenizer_exceptions problem with Persian

[mosynaq]

I am trying to train Persian to spaCy. One of the problems is here: in tokenizer_exceptions.py, spaCy expects concatenation of two orths, form the word itself like do + n't = don't, but for Persian, this expectation is not valid for some cases.
For example, the verb "بر نخواهد گشت" ( = s/he will not return), is made up of "بر" + "نـ" + "خواهد گشت".
("نـ" negates a Persian verb. Most of the times the negation thing comes in the beginning, but in some, cases like this one, it comes in between. )
As you can see you cannot simply concatenate orthes to form the full form.

Should spaCy expectation be changed? Or should I do something?

My Environment

• spaCy version: 2.0.5
• Platform: Linux-4.11.0-14-generic-x86_64-with-LinuxMint-18.2-sonya
• Models: en, fa
• Python version: 3.5.2

[honnibal]

It's true that we don't really have a good solution to this for Persian, Arabic and other similar languages. I'm still not positive what the best strategy should be.