Due to the growing amount of literature, the need for automated extraction of relations between genes, metabolites and phenotypes from natural language text has steadily been increasing over the last years. Several algorithms for extracting knowledge from natural language texts have been developed and improved. This work aimed at the development of a broad scale text mining system covering a multitude of relation as well as entity types. The resulting text mining system EXCERBT was developed, optimized and evaluated in hindsight on practical usability rather than on optimized precision or recall values for a singular relation extraction task. EXCERBT is a dictionary based text mining system based on Semantic Role Labeling in combination with cooccurrence. The system allows semantic queries for genes causing a certain phenotype or miRNAs inhibiting a certain gene. In addition, EXCERBT comprises a new approach for automatically generating biomedical lexica by means of Semantic Role Labeling.