从范式嬗变看语言学与人工智能的融合路径

詹卫东

语言战略研究 ›› 2026, Vol. 11 ›› Issue (3) : 41-52.

欢迎访问《语言战略研究》官方网站!今天是
语言战略研究 ›› 2026, Vol. 11 ›› Issue (3) : 41-52. DOI: 10.19689/j.cnki.cn10-1361/h.20260303
专题研究

从范式嬗变看语言学与人工智能的融合路径

  • 詹卫东
作者信息 +

Rethinking the Integration of Linguistics and Artificial Intelligence Through Paradigm Shifts

  • Zhan Weidong
Author information +
文章历史 +

摘要

从规范语言学、比较语言学、结构语言学到生成语言学,语言学跨越千年的范式变迁,终极目标是为人脑内部语言建模。语言学者虽已积累大量微观语言学成果,却难以由量变引起质变,距建构整体语言模型仍有巨大鸿沟。自1950年代至今,人工智能技术范式从符号主义、经验主义发展到联结主义,生成式大语言模型已经可以依靠强大算力和海量数据,对人类语言进行全量建模。但在挖掘表层分布没有充分表达的深层认知规律方面,还无法达到母语者水平。语言学者可以基于自身的语言学洞察力,充分挖掘微观语言现象中蕴含的深层语义问题,把语言知识转化为高质量语言数据,帮助提高人工智能的语言能力。同时利用人工智能技术,在观察充分、描写充分、解释充分的基础上,进一步实现“生成充分”。如何由理论研究成果驱动,由纯手工到半自动再到全自动地生成能与人工智能直接交互的语言数据,是人工智能时代语言学者要认真思考的问题。

Abstract

Across its long intellectual history, from prescriptivism and descriptivism to structuralism and generative grammar, linguistics has the ultimate goal of modelling the internal language faculty of the human mind. Despite the accumulation of extensive body of micro-level findings, a substantial gap persists between such fragmented insights and the construction of a holistic model of human language as a cognitive system. Since the 1950s, the dominant paradigm trajectory in artificial intelligence has evolved from symbolism through empiricism to connectionism. Although contemporary large language models (LLMs), powered by leveraging massive computation and vast datasets, can model human language at unprecedented computational power, they remain limited in capturing the deep cognitive regularities that are not fully encoded in surface distributions and therefore still fall short of native-speaker competence in extracting and internalizing deeper cognitive and semantic regularities that are weakly expressed in surface distributions alone. In this context, it opens a critical space for collaboration. Linguists, drawing on their theoretical sensitivity to subtle semantic, pragmatic, and structural phenomena, can contribute by leveraging domain-specific insight to uncover the deep semantic issues embedded in micro-level linguistic phenomena and to transform linguistic knowledge into high-quality, structured data for improving AI’s linguistic capabilities. At the same time, AI offers linguistics the possibility of moving beyond observational, descriptive, and explanatory adequacy toward generative adequacy. Consequently, the automated transformation of theoretical research into interactive linguistic data—evolving from manual to fully autonomous processes—constitutes a central challenge for linguists in the AI era, and addressing this challenge is essential if linguistics is to play a constitutive role in the next stage of language modelling.

关键词

生成式人工智能 / 形式文法 / 大语言模型 / 深度学习

Key words

generative artificial intelligence / formal grammar / large language models / deep learning

引用本文

导出引用
詹卫东. 从范式嬗变看语言学与人工智能的融合路径[J]. 语言战略研究. 2026, 11(3): 41-52 https://doi.org/10.19689/j.cnki.cn10-1361/h.20260303
Zhan Weidong. Rethinking the Integration of Linguistics and Artificial Intelligence Through Paradigm Shifts[J]. Chinese Journal of Language Policy and Planning. 2026, 11(3): 41-52 https://doi.org/10.19689/j.cnki.cn10-1361/h.20260303
中图分类号: H002   

基金

教育部人文社会科学重点研究基地重大项目“面向机器语言能力评测的综合型语言知识库研究”(22JJD740004)。本研究同时得到多媒体信息处理全国重点实验室开放课题基金(SKLMIP‒KF‒2025‒01)的支持,特此致谢!

Accesses

Citation

Detail

段落导航
相关文章

/