The automatic processing of the Arabic language is a growing discipline, in which one sees more and more research and technologies to examine the specificities of this language and to propose tools necessary to the development of its automatic processing. The old techniques of rooting have limits that weaken the process of root extraction. In this article, the author proposes a new approach to rooting based on two finite state automata. The technique proposed is based on finite state automata in the root extraction process, with the aim of minimizing the error rate and ambiguity, usually due to the removal of the affixes. The author is currently focusing on the development and improvement of the rooting technique while trying to overcome the various problems encountered. The author is working on the compilation of a corpus of evaluation which will allow him to evaluate and compare their approach to others