Morphological analysis is an essential step in many natural language processing applications. Indeed, information retrieval and processing algorithms often use a morphological analyzer as a text preprocessing step. The rich morphology of the Arabic language, combined with the frequent absence of diacritical marks in texts, make morphological analysis particularly complex. Numerous morphological analyzers have been developed for the Arabic language using machine learning techniques such as Hidden Markov Models and support vector machines. In this article, we present a new Arabic lemmatizer that combines deep learning and out-of-context morphological analysis using the Alkhalil Morpho Sys analyzer. The tests carried out demonstrate the effectiveness of this approach, and an accuracy of the order of 98% was achieved in the test set.