Bounded accuracy and loss "plateau"

[gastodler]

Hi,

I'm training a custom SAR in Arabic.

I've tried many directions, but I'm stuck where, and my loss get into a "plateau", and my accuracy is bounded around 0.6 (word level).

I'd like to know where my problem is:

1. In the training set
2. in my dict
3. in my learning rate
4. somewhere else / all together

I've tried to check specific areas. for example, I have an image of ("&") symbol in my training set, but when I check it using one of my trained models -> they don't recognize it right. my dict has this symbol in it ("&"), but I'm still not sure that it says my dict is "correct". I have some similar issues with numbers and other symbols.

Any help would be appreciated.