大语言模型与语言学研究：演进脉络、应用图景与反思

Large Language Models and Linguistic Research: Evolutionary Trajectory, Application Landscape, and Reflections

摘要: 大语言模型的发展深刻重塑语言学研究范式。立足于语言学学科本位，本文梳理近十五年大语言模型由词嵌入、Transformer架构到生成式爆发的演进脉络，从语言本体与应用层面考察其在语义、句法、语用、语言教育、机器翻译、语料库语言学中的应用图景。研究表明，现有交叉研究多侧重技术测评，普遍将模型视作单纯工具，忽视二者双向互动。当前研究存在模型本身语言学缺陷突出、重工具轻理论、传统理论解释力遭遇挑战、方法论困惑与语言治理风险并存等问题。据此，本文提出推动理论双向融合、推进汉语及低资源语言本土化、构建人机协同研究范式、完善语言伦理治理的优化路径。语言学应当主动适配数据驱动范式，在技术与人文交叉地带实现理论与方法的创新发展。

Abstract: The development of large language models has profoundly reshaped the research paradigms in linguistics. Grounded in the disciplinary framework of linguistics, this paper traces the evolutionary trajectory of large language models over the past fifteen years—spanning from word embeddings and the Transformer architecture to their generative surge—and examines their applications across the domains of linguistic ontology and practice, including semantics, syntax, pragmatics, language education, machine translation, and corpus linguistics. The study reveals that much of the current interdisciplinary research focuses primarily on technical evaluation, often treating language models as mere tools while overlooking the bidirectional interaction between models and linguistic theory. Current research faces several challenges: prominent linguistic limitations within the models themselves, an overemphasis on tools at the expense of theory, the declining explanatory power of traditional linguistic theories, and the coexistence of methodological uncertainties alongside risks associated with language governance. In response, this paper proposes an optimal pathway to foster bidirectional integration of theory and practice, advance the localization of research approaches for Chinese and low-resource languages, establish a human-machine collaborative research paradigm, and enhance language ethics governance. Linguistics should proactively adapt to data-driven paradigms and drive innovative theoretical and methodological developments at the intersection of technology and the humanities.