Automated Prediction of Medieval Arabic Diacritics

Abstract

This study uses a character level neural machine translation approach trained on a long short-term memory-based bi-directional recurrent neural network architecture for diacritization of Medieval Arabic. The results improve from the online tool used as a baseline. A diacritization model have been published openly through an easy to use Python package available on PyPi and Zenodo. We have found that context size should be considered when optimizing a feasible prediction model.

Tasks

Reproductions