We describe ongoing work in semi-automatic annotating corpus, with the goal to answer why-question in question answering system and give a construction of the coherent tree for text summarization. In this paper we present annotation schemas for identifying the discourse relations that hold between the parts of text as well as the particular textual of span that are related via the discourse relation. Furthermore, we address several tasks in building the annotated corpus in discourse level, namely creating annotated guidelines, ensuring annotation accuracy and evaluating.