One of the challenges of today's information society is to cope with the immense amount of electronic data, as for instance provided by the internet. It would already be helpful to have a tool, which allows verifying a hypothesis from the information given in a natural-language text. Since this requires the tool to precisely "understand" what is written in the text, it seems unavoidable to perform a deep linguistic analysis of the text. This leads to an abstract representation of the text. The frame-structures provided by the Berkeley FrameNet Project seem a natural choice for this representation. This thesis will show how the frame-representation can be converted to PROLOG-like rules. By these rules, it will be possible to reason on the representation and finally to derive or refute the hypothesis. Thus, this thesis will provide rule-based semantics for the FrameNet-representatio