Serious games are often used in education as they provide an excellent opportunity for learning. This paper presents our proposition of a system capable of generating adaptive learning scenarios for serious games. Thus, we have presented the architecture of this system. The domain knowledge is organized in three layers. These layers include the domain concepts, the pedagogical resources and the serious game resources. The approach we propose is capable to keep in account the specificities of serious games, the user profile and his/her goals while generating scenarios. Furthermore, it also uses the interaction traces as knowledge sources in the adaptation process. To test the applicability of our system we are working on the Project CLES (Cognitive and Linguistic Element Stimulation). This project targets producing serious games for users with cognitive disabilities. We’ve presented an example of how our system will generate a pedagogical scenario.