Empirical Evidence: The Semantic Structural Evolution from "Ordinary Novelty" to "Literary Classic"

In order to break through the limitations of existing research (Yang et al. 2025) that is limited to the generation of metaphors in ordinary subjects, and to address the theoretical gaps in the use of metaphor calculation models to explain higher-order creativity, this section aims to reveal the deep computational rules of metaphor evolution from "everyday expressions" to "literary classics" by introducing a new computational framework. 3.1 Corpus Construction and Standardization Processing We have constructed a comparative corpus consisting of three levels: Firstly, Bad Metaphors: selected from the metaphorical corpus generated by college students in the early stage of Yang et al. (2025). The metaphorical creativity score in this corpus was evaluated by two trained experts strictly based on three core indicators: novelty, remoteness, and cleverness (Silvia&Beaty 2012), using a 1-5 point Likert scale. This study specifically selected 78 corpora with extremely low scores (≤ 1.5); Secondly, Good Metaphors: Also from the aforementioned corpus, 47 excellent creative metaphors with extremely high scores (≥ 4.5) under the same rating criteria were selected; Thirdly, Classic Metaphors in Literature: A total of 47 famous metaphors that have been systematically extracted from famous works of Chinese and foreign scholars and thinkers, tested over time, and standardized in form. Specifically, this study selects language materials from literary classics (such as Shakespeare and Su Shi), philosophical works (such as Plato and Laozi), and political discourse (such as Marx and Churchill) based on the three principles of "classicism", "cross domain mapping salience", and "structural integrity". The selected corpus covers four core semantic fields: (1) time, history, and memory; (2) Life, destiny, and existence; (3) Emotional and psychological states; (4) Social and political ideology. The corpus sources widely cover Chinese classical poetry, modern literary classics, as well as Western philosophical and literary classics. In order to standardize the calculation, foreign language materials are accurately translated into Chinese by professionals to eliminate the interference of cross linguistic morphological differences on the extraction of semantic vectors from large models. In order to ensure strict inter group comparability of the three levels of corpus in computational analysis, this study invited two linguistics PhDs to reconstruct all classic corpora into an explicit propositional structure of "ontology metaphor common features" (i.e., "T like B, because they are all F") that is completely consistent with the laboratory collected corpora. Two experts independently extracted the "key common feature (F)" that best conveys the original author's intention in the context of the original text and conducted cross checking. This rigorous standardization procedure lays an objective data foundation for extracting high-dimensional semantic vectors in the future.