Zero-shot Cross-lingual EAE has garnered significant interests from the community because it could minimize the need for extensive data annotation to identify the roles of the arguments within a specific event. Some prior works point out that syntactic structures could be regarded as the language-independent features and those methods have achieved promising performance. However, sometimes even the sentences in different languages express the same meaning, the syntactic parsing results are quite different. To alleviate this problem, we find the semantic information is rarely considered, which could be considered as another language-independent features and provides more consistent parsing results across languages. To this end, in this paper, we propose the Semantic-Syntactic Driven framework (S2D) for incorporating the semantic and syntactic information simultaneously. Specifically, we design a language-independent dual-prefix constructor module to handle the semantic and syntactic discrepancies between source and target languages. Besides, we introduce a semantic role labeling source generator module to expand the source language dataset with semantic features. The experimental results illustrate that S2D surpasses the state-of-the-art methods by 3.4% and 5.3% in terms of average F1-score on the ACE-2005 and ERE datasets, respectively.