Event management and response generation are two essential aspects of systems for ambient intelligence. This work proposes handling these issues through the use of a semantic model for ambient intelligence which, under the umbrella of a philosophical and common-sense optic, describes what actions and events are, how they are connected, and how computational systems should think about their meaning. This model entails an approach with which to both reason about and model context events and generate behavioral responses to those events, when necessary. The model supports this ad hoc response generation by automatically composing services when those which are available do not meet the expected functionality (without requesting user intervention). An evaluation methodology is presented and illustrated with a case scenario, in which synthetic data has been generated to emulate events and analyze the system response. The evaluation of the system response is carried out on the basis of a vector of attributes.