The knowledge contained in the growing number of scientific digital publications, particularly over the internet creates new demands for intelligent retrieval mechanisms. One basic approach in support of such retrieval mechanisms is the generation of semantic annotation, based on ontologies describing both the field and the structure of the texts themselves. Many current approaches use statistical methods similar to the ones employed by Google to find correlations within the texts. This approach neglects the additional information provided in the upper ontology used by the author. mArachna, however, is based on natural language processing techniques, taking advantage of characteristic linguistic structures defined by the language used in mathematical texts. It stores the extracted knowledge in a knowledge base, creating a low-level ontology of mathematics and mapping this ontology onto the structure of the knowledge base. The following article gives an overview over the concepts and technical implementation of the mArachna prototype.