大语言模型视角下汉语动物歇后语构式语义及隐喻机制研究

A Study on Constructional Semantics and Metaphorical Mechanisms of Chinese Animal Two-Part Allegorical Sayings from the Perspective of Large Language Models

摘要: 汉语动物歇后语具有构式的合法地位，其构式义具有高度的整体性和强烈的语用功能，往往无法单独通过前件、后件或既有构式直接推知，兼具认知构式与文化构式性质。当下，以ChatGPT、DeepSeek等为代表的大语言模型虽广泛应用于自然语言处理领域，但它们有时仍无法像人类一样富有体验性地准确理解汉语动物歇后语的构式义。本文基于对动物歇后语句法结构特征的分析，归纳出其语义指向的三层构式体系，即实例人物层、模式事件层与抽象理据层；并进一步揭示出两种隐喻机制，即后件单独充当源域和前后件联合充当源域。研究力求为提升大语言模型的语义理解与文本生成能力提供新的分析路径。

Abstract: Chinese animal two-part allegorical sayings possess the legitimate status of constructions. Their constructional meanings are highly holistic and functionally pragmatic, which usually cannot be directly inferred from the former part, the latter part, or existing constructions alone. They bear the properties of both cognitive constructions and cultural constructions. At present, large language models represented by ChatGPT, DeepSeek and others have been widely applied in the field of natural language processing. However, they still sometimes fail to understand the constructional meanings of Chinese animal two-part allegorical sayings experientially and accurately as humans do. Based on an analysis of the syntactic structural features of Chinese animal two-part allegorical sayings, this study identifies a three-tier construction system of semantic orientation, consisting of the instance-character layer, the pattern-event layer, and the abstract-motivation layer. Furthermore, it reveals two types of metaphorical mechanisms: the latter part independently serving as the source domain and the joint former and latter parts serving as the source domain. The research aims to offer a new analytical pathway for enhancing the semantic comprehension and text generation capabilities of large language models.