Answers are grounded in this paper and cite their sources.
Recent advancements in generative AI technology have increased the role of AI in academic research. However, the capability of different generative AIs in analyzing metaphors has not been adequately investigated. The purpose of this paper is to compare the capability of ChatGPT, Gemini, and DeepSeek in analyzing fire metaphors in online discourse. Data were based on the iWeb Corpus, and 100 concordance lines of three verbs that potentially instantiated fire metaphors, namely, “spark,” “fuel,” and “dampen,” were collected for analysis by the human researcher. The data were then analyzed by each of the generative AIs. The analyses were compared against those of the researcher. Cohen’s Kappa was used to determine the intercoder agreement. The analysis reveals 14 target domains associated with the fire metaphors. The results indicate that there is substantial agreement between the researcher and each of the generative AIs. However, a closer look at the analysis of individual words reveals variation from moderate to almost perfect agreement, with DeepSeek scoring the highest. The implications of the roles of AI in metaphor research are discussed.
No takes yet. Share an insight, caveat, or question.
Kheovichai et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: