I have question, USE for Arabic works for dialect ?? or only modern standard arabic?
The Universal Sentence Encoder Arabic is a standard pre-trained model. If you are looking to use for dialect as per your usecase then you should re-train that model with your dataset on top of the pre-trained model.
hub.KerasLayer(pre-trained-model,trainable=True,....)
For more details please refer here.
Thank you!