Neural Machine Translation from Jordanian Dialect to Modern Standard Arabic

The development of cultures and societies all over the world was the first reason for the emergence of many different languages and dialects that differ from each other based on the geographical location of these communities, whether in the Arab countries or Western or other parts of the world. Due to these differences, there is a need to translate these dialects between each other to facilitate their understanding and handling by people who will use them from other communities. The tremendous technological advancement and the flourishing of the era of Deep Learning, has led to the emergence of so-called neural machine translation (NMT), which has significantly facilitated the translation process compared to other methods. In this paper, we present a framework to translate the Jordanian dialect into Modern Standard Arabic (MSA) using Deep Learning, in particular, the RNN encoder-decoder model, which provided good results on the level of our manually created dataset. The conducted experiments using this model were divided into two parts: word level and sentence level, and the results were as follows: loss equals 0.8 and accuracy equals 91.3% when using the model for word to word translation; and loss value equals 3.33 and accuracy equals 63.2% when using the model for sentence translation. These are very encouraging results in this largely unexplored topic.