Accent recognition refers to the problem of inferring the native language of a speaker from his foreign-accented speech. Differences in accent are due to both articulation and prosodic characteristics. The automatic identification of foreign accents is valuable for different speech systems, such as speech recognition, speaker identification or voice conversion. This paper aims to identify the native languages of non-native English speakers from different countries in the Arabic region: the researchers choose Saudi Arabia to represent the eastern Arabic region in Asia, Egypt to represent the eastern Arabic region in Africa and Tunisia to represent the western Arabic region in Africa. In this research, reinforcement learning (RL), a sub-branch of machine learning, and artificial neural network will be used as an intelligent method to classify and predict the speech. The aim is to train a neural network to automatically detect speech accents. Then the researchers develop a hybrid multi-agent RL algorithm that takes advantage from the multi-agent communication and cooperative agents on the language detection process. Hence, the aim is to help sociolinguists and discourse analysts. As for the Saudi context, this study will be very useful on resolving e-learning issues such as linguistic problems of students.