Comparative Analysis of Business English and Everyday English Translation Stylistic Features Based on Keras Neural Networks

Abstract

Based on the formal definition of semantic language and specific natural language, the research adopts the Word2vec model to map words into a low-dimensional dense space and performs semantic modeling by unsupervised learning through a large amount of unlabeled data to realize the vectorized representation of translated text. For the vectorized text, the text features are combined and filtered based on the TextCNN network, and the TextCNN network is implemented under the Keras framework. Based on the TextCNN network of Keras, the stylistic features of Business English and Daily English are compared at three levels: lexical, syntactic and discourse. Business English was rated 23.6 points higher in terms of conciseness of diction compared to everyday English, and 28.7 points improved the modesty of diction compared to everyday English. This study furthers the improvement of various English translation types.