We describe the adaptation for Arabic of the grammar-based MedSLT medical speech system. The system supports simple medical diagnosis questions about headaches using vocabulary of 322 words. We show that the MedSLT architecture based on motivated general grammars produces very good results, with a limited effort. Based on the grammars for other languages covered by the system, it is in fact very easy to develop an Arabic grammar and to specialize it efficiently for the different system tasks. In this paper, we focus on generation.