Author(s): Abdelhamid. Atassi, Ikram El Azami
In this paper, we will demonstrate how we can automatically generate a new poem in Arabic language from a dataset containing a number of poems written by renowned Arabic poets and writers based on classical Arabic language and from colloquial Arabic. A comparison was also made between the different LSTM (Long Short-Term Memory), BiLSTM (Bi-directional Long Short-Term Memory) and GRU (Gated recurrent unit) generation networks in order to obtain a satisfactory result, keeping in mind the main one. The difference between a text in Arabic language and another in English or French is the orientation of the text, which is from right to left. For this reason, we tried to use the Bi-LSTM to have the added value of this network, but the latter did not lead to correct results neither at the sentence level nor at the word level, contrary to the GRU to give good results in terms of performance.