The language features of English translation are complex and diverse, including semantic ambiguity, syntactic diversity, and stylistic differences, which often make it difficult to accurately capture the inherent connections between language features and affect effective collection. Therefore, in order to improve the effectiveness of data collection, a method for intelligent collection of English translation language features based on graph neural networks is studied. Through multidimensional analysis theory and principles, reveal the changing forms of language features in English translation from semantic, syntactic, stylistic, and other forms; Choose to use similarity metrics to determine the degree of correlation between different language features, in order to more accurately identify the intrinsic connections between language features and provide more valuable feature information for subsequent translation and language processing tasks. Based on the similarity judgment results, classify the types of English translation language features; Build a graph structure for translating text, use it as input to establish a model using graph neural networks, capture the complex relationships between nodes in the graph structure, and better understand the semantic, syntactic, and stylistic features in the translated text, achieving accurate intelligent collection of English translation language features. The results indicate that the research method effectively classifies English translated texts, collects language features within different translated texts, and has practical value.