Achieving the robust and universal semantic representation for action description remains the key challenge in natural language understanding. Current approaches often struggle to capture the complexity of human actions, leading to limited representations. To address this challenge, we propose a