A computational model of poetry appreciation based on a spreading activation network

Abstract

In studies on aesthetic appreciation, poetry has received less attention than other art forms. Furthermore, few attempts have been made to model the underlying cognitive processes, which leaves computational approaches insufficiently explored. In this study, to explore the cognitive process of poetry appreciation, we propose a computational model of poetry appreciation based on a spreading activation network. According to the Aesthetic Association Principle, poetry reading comprehension involves the process of evoking semantically related concepts from the words in the poem. We model this process as spreading activation within a semantic network. We model the aesthetic effect using cognitive load and its resolution based on incongruity resolution theory. Specifically, we quantify the diversity of activation in the network using Shannon entropy and calculate the degrees of incongruity resolution and unresolved load from entropy differences. We tested the computational model using a linear mixed model, with simulation results as the independent variables, human ratings of aesthetic appeal as the dependent variable, and participants as the random effect. The analysis demonstrated that the degree of incongruity resolution had a significant positive effect and the degree of unresolved load had a significant negative effect on aesthetic appeal. This finding is consistent with incongruity resolution theory, thus providing evidence for the validity of the model and suggesting its potential as a comprehensive model of poetry appreciation.