Predictive Natural Language Processing Analysis Applied to Arabic

This paper aims at applying Recurrent Neural Networks (RNN) to Natural Language Processing using old texts such as sacred books. The objective of this work is to generate a text that follows the same structure and semantic of the studied corpus written in Arabic. This is done through various networks such as vanilla RNN, LSTM (Long Short Term Memory) and GRU (Gated Recurrent Unit). After training, the quality of the generated text is measured using BLEU scores. Besides, an exploratory analysis of the corpus is presented to exhibit some interesting findings at the words and chapters level.