Designing serious games or serious game activities requires mapping the educational elements and the game elements. This mapping is mainly addressed from a high-level game design perspective. Moreover, low-level mapping methods are generally domain-specific. Our aim is to address this problem, at an algorithmic level, in the context of activity generation (i.e., automatic creation of activities) for declarative knowledge training. This paper presents a generic modelling approach of questioned facts and gameplays, and an algorithm for the automatic and domain-independent generation of various gameplays for training purposes. The approach has been applied to multiplication tables training.