Code-switching is the choice of a language, a variant of using multiple languages in the same conversation. Broadly speaking, code-switching refers to adjusting one's language style, appearance, behavior, and expression in order to improve the comfort of others in exchange for fair treatment, quality service, and employment opportunities. "Besieged City" is considered a masterpiece of 20(th)-century China. From the data point of view, this work has a total of 110 code shifts, but there are many studies on this language phenomenon, but none of them involve the perspective of register. However, language translation research based on register theory is of great significance. It is generally believed that the human brain's thinking is divided into three basic ways: abstract (logical) thinking, image (intuitive) thinking, and inspiration (awareness) thinking. Artificial neural networks are the second way to simulate human thinking. Therefore, this paper proposes research on cognitive feature extraction of pun code-switching based on a neural network optimization algorithm. It mainly introduces code-switching under cognitive language and also briefly analyzes code-switching and speech feature extraction and uses a neural network optimization algorithm to conduct an in-depth analysis of code-switching. Finally, in the experimental part, the experimental analysis of the famous novel "Besieged City" is carried out, the application of 89 language code-switching in the text is deeply analyzed, and the data analysis of its three variables is carried out from the perspective of register. The experimental results show that: in novels, there are two types of code-switching: preparation and improvisation. 24 code-switches are prepared, accounting for 26.9%, and 10 code-switches are improvisation, accounting for 8.9%. As for the verbal code-change, there are both preparatory and impromptu ones. 36 code-switching cases were improvised, accounting for 40.4%, and 19 code-switching cases were prepared, accounting for 21.3%. The analysis also confirms that the more formal the text, the less linguistic transformation it contains.
