For most events trigger word extraction failed to make full use of syntactic relations and semantic information of problem, put forward the fusion syntax and semantic information of multi-dimensional Chinese emergency trigger word extraction. First of all, using the pre-training language model Bidirectional Encoder Representations from Transformers with Whole Word Masking (BERT-wmm) for embedded sentence initial vector Character and Word vector; Then, through the Graph Transformation Multi order Graph Attention Network to generate a new dependent relationship between arc dependency Graph, to capture the words more jump dependencies and potential interactions information between words, and enhance the information expressing ability of words; At the same time, the use of double layer additional attention mechanism based Bidirectional gated Recurrent neural network with sentence level information and by the number of interdependence sentence level of Chapter level information, to extract the key information of multidimensional semantic information; Finally, the word, word, sentence and chapter level information fusion using conditions with the airport complete event trigger word extraction. In CEC Chinese data set on the experimental results show that this method can effectively enhance the Chinese emergency trigger word extraction effect.