We present the conceptual and technological foundations of a distributed natural language interface employing a graph-based parsing approach. The parsing model developed in this thesis generates a semantic representation of a natural language query in a 3-staged, transition-based process using probabilistic patterns. The semantic representation of a natural language query is modeled in terms of a graph, which represents entities as nodes connected by edges representing relations between entities. The presented system architecture provides the concept of a natural language interface that is both independent in terms of the included vocabularies for parsing the syntax and semantics of the input query, as well as the knowledge sources that are consulted for retrieving search results. This functionality is achieved by modularizing the system's components, addressing external data sources by flexible modules which can be modified at runtime. We evaluate the system's performance by testing the accuracy of the syntactic parser, the precision of the retrieved search results as well as the speed of the prototype.