Based on the method of cognitive mapping, this paper systematically organizes the evolution of Yuzhong dialect vocabulary in different historical stages, revealing its important role in the evolution of the Chinese language. The global statistical information of the corpus is utilized to train the word vectors, which are input into the recurrent neural network as feature vectors to enhance the model’s ability to learn long-distance information. Perplexity is defined as a metric for evaluating the performance of a language model, the inverse of the geometric mean probability assigned to each word by the model given a known sequence of words in the test set. Subsequently, based on this metric, a corresponding model was built to comprehensively evaluate the performance of the language model, which provides insights into the functions and expressions of Yuzhong dialect words in communication. Among them, the number of unsubstituted nouns, such as weather, is 41, and the color of the sky is 67, showing the rhyme and stability of Yuzhong dialect vocabulary in specific domains.