A Study of English Rhetorical Strategies Based on Neurosemantics

In recent years, with the performance improvement of deep neural network models, neural semantic parsing has entered a new stage and has been widely used in many fields such as classification tasks, speech recognition, linguistic annotation and syntactic analysis. In this paper, we design a strategy for recognizing and parsing English rhetorical devices using the neural semantic parsing method. In this paper, we first use the Skip-gram model to train word vectors, and then based on the convolutional neural network model to recognize the platitudes, metaphors, and similes in English text, and apply BiLSTM-Att model to acknowledge the English quoted rhetorics. The prediction accuracies of the CNN recognition model on platitudes, metaphors, and similes are 91.7%, 92.5%, and 90.2%, respectively, greater than 90%, indicating that the model can successfully recognize the platitudes, metaphors, and similes. The model can recognize platitudes, metaphors, and anthropomorphic rhetorical devices. Compared with the CNN+BiLSTM model, the precision, recall, and F1 value of the BiLSTM-Att model are improved by 0.22%, 11.44%, and 7.09%, respectively. Its recognition accuracy of English quotations under different similarity thresholds is very high, with 94% when the similarity reaches 50% and even 98% when the similarity reaches 90%. This suggests that the present paper BiLSTM-Att English quote rhetorical recognition model is adequate. This paper’s neural semantic-based English rhetoric strategy is paving the way for the development of natural language processing technology.