GRAAL-Research/deepparse

training a model for non Latin language

maherr13 opened this issue · 5 comments

maherr13 commented

Hello

I'm trying to use your package for training address parser for arabic (different letters from proposed languages), do you offer this feature ? if so would you guide me how to achieve this.

Thanks

github-actions commented

Thank you for you interest in improving Deepparse.

MAYAS3 commented

Hello @maherr13,

Deepparse's models are multilingual, so they should support the arabic language as long as the dataset is defined properly.

You can take a look at our retraining tutorial (https://deepparse.org/examples/fine_tuning.html) and our training guide (https://deepparse.org/training_guide.html) for more information on how to train the models on your data.

Let me know if you have any questions!

maherr13 commented

image
I read this part in the paper, do you think it will work with Arabic letters?

davebulaval commented

@maherr13 it works in Japanese, Chinese and other non-Latin languages. I am pretty sure it will work.

MAYAS3 commented

@maherr13 I'm closing this issue. Feel free to experiment with our models and let me know if you encounter any issues.