The Integration of Large Language Models and Linguistic Theory: Theories and Evaluation

Abstract

This paper focuses on existing research at the intersection of large language models and metaphor, offering a systematic review organized around two core issues: theoretical integration and capability evaluation. First, it examines how linguistic theories, including Conceptual Metaphor Theory and Critical Metaphor Analysis, have been integrated with large language models and analyzes their contributions to model performance. Second, from a linguistic perspective, the paper defines the metaphor-processing capabilities of large language models along four dimensions: metaphor identification, interpretation, generation, and tool use, and evaluates model performance across these dimensions. Based on its examination of these two core issues, the paper argues that future research should broaden and deepen the integration of linguistic theory and computational methods, while further exploring approaches to evaluating and enhancing the linguistic expertise of large language models.