Document-level event argument extraction (DEAE) aims to recognize event arguments spreading across multiple sentences and their corresponding roles. DEAE is a challenging but indispensable subtask in general document-level event extraction. Existing methods are not effective due to two challenges of this task: (a) traditional extraction methods cannot assign multiple labels to a token, so these methods are unable to identify two nested arguments belonging to different roles; (b) previous methods are weak in encoding the document-level sequence and they have never modeled the relevance between different roles. In this paper, we propose a Role-aware Interactive Pointer Labeling Network (Ripl) to solve these two challenges. For challenge (a), we construct specific document representations for each role and propose a novel interactive pointer labeling strategy to extract nested arguments. For challenge (b), we introduce role prior knowledge into the original document, make the representation of each word in document captures the semantic information of the role and model the semantic relevance among the roles. Experiments on the MUC-4 event extraction dataset show that Ripl outperforms the previous state-of-the-art model by 5.9 F1 on the head noun match metric. Then we demonstrate the effectiveness and interpretability of our model with extensive experiments.