A classical activity theory has for long been established as a de-facto framework on which many human activity models and recognition algorithms are based. We observed several limitations to the classical activity theory, mostly attributed to its inability to discern among seemingly similar concepts and artifacts. We discuss these limitations and introduce a generic activity framework that addresses these issues. We provide a comparative analysis to quantify the effect of our proposed framework on the accuracy of activity recognition using a benchmark of activities of daily living. We report significant improvement of the recognition accuracy. © 2011 ICST Institute for Computer Science, Social Informatics and Telecommunications Engineering.