In recent years several efforts have been devoted by researchers in the Requirements Engineering community to the development of methodologies for supporting designers during requirements elicitation, modeling, and analysis. However, these methodologies often lack tool support to facilitate their application in practice and encourage companies to adopt them. In this paper, we present our experience in the application of methods for the transformation of requirements specifications expressed in natural language into semi-structured specifications. More specifically, we apply a lightweight method for extracting requirements from system descriptions in natural language to support the Secure Tropos methodology during requirements elicitation phase. Our proposal is based on Cerno, a semantic annotation environment, which uses high-speed context-free robust parsing combined with simple word search. To evaluate our proposal, we discuss its application to the requirements elicitation process followed in the course of a European project on four industrial case studies.