Event argument extraction is a challenging subtask of event extraction. It aims to identify event argument spans and predict their roles. Recent studies often formulate this task as a question-answering or sequence generation task. However, the design of questions or prompts and their templates often requires substantial manual effort. Additionally, most existing methods fail to effectively utilize rich external knowledge. To address these issues, we propose CFNEAE (Chinese FrameNet Event Argument Extraction), a framework leveraging Chinese FrameNet as an external knowledge base. CFNEAE utilizes prior argument role definitions as prompts for Chinese argument extraction. We also introduce bi-cross attention between the context and argument roles, modeling the context and argument roles from two directions. Furthermore, a graph attention network is employed to capture underlying relations among argument roles. Finally, a multi-layer pointer network is used for decoding, addressing the challenge of overlapping roles. Extensive experimental results demonstrate that CFNEAE achieves state-of-the-art performance on the DuEE and FewFC Chinese event extraction datasets. Additional experiments demonstrate CFNEAE’s effectiveness in extracting event arguments with multiple and overlapping roles.