In this paper, a multimodal phrase corpus is created, enabling phrases to be searched automatically. The phrase structure is constructed through phrase centroids, and multilevel linguistic measurements are obtained for vocabulary, syntax, and parts of speech styles. A joint feature analysis model for the automatic conversion of English-Chinese machine translation is constructed through distributed feature fusion and joint parameter analysis. Linear statistical features and feature analysis detection are utilized to compute the optimal solution for English-Chinese machine automatic translation text. The graph embedding technique was used to pretrain the knowledge graph and obtain information about the network structure. Combined with the graph convolution technique, deep semantic information mining was carried out to achieve efficient content matching and cultural dissemination. It has been proved that the method of this paper has an AUC value of 0.921 and a satisfaction rate between 80% and 95% in English-translated movie data. The proposed method can accurately translate information from English to foreign propaganda content, which further promotes the development of international cultural exchange and communication.