Best practice for Arabic recognition training - how to create a dictionary

[Hegelim]

Dear Community,

I have an urgent question about Arabic recognition training and since there aren't many related information online, I am here asking for help.

Let's say I want to train an Arabic recognition model, what's the best practice when creating a customized Arabic dictionary?
Now, there are several things that make it challenging:

1. Arabic letters change their shapes depending on their locations in the word, for example, the letter alif has 4 forms and each one has a unicode glyph. Should I include all possible shapes of it in the dictionary or should I just include a single letter in the alphabet?
2. follow-up on 1, if I only include a single letter, then how is the model trained such that it can recognize different shapes of the same letter? It sounds like a 1-to-many mapping, can the model do that?
3. Arabic is cursive, that means when joining letters together, they merge together, which is called ligature. How can I take this into account when creating the dictionary?
4. What's the order of the MMOCR recognition? Because Arabic is a right-to-left language, and if MMOCR reads texts from left to right, should I be concerned and are there any files that I should change?