This paper describes the development of a relational discourse parsing architecture for text documents of a complex text type, namely scientific articles. To achieve this goal, several different linguistic knowledge sources and auxiliary analyses on different linguistic levels are necessary.