大语言模型与生成语言学的范式对比  

A Comparison of the Paradigms in Large Language Models and the Generative Linguistics

作　　者：王晨 WANG Chen(School of Foreign Studies,Nanjing University of Science and Technology,Nanjing,Jiangsu,210094)

机构地区：[1]南京理工大学外国语学院,江苏南京210094

出　　处：《南京理工大学学报（社会科学版）》2024年第4期61-69,共9页Journal of Nanjing University of Science and Technology：Social Sciences

基　　金：南京理工大学自主科研项目“跨语言视角下方位系统的生成语法研究”(30923010411)。

摘　　要：算力的提升和数据库的规模扩大使得现代大语言模型在自然语言处理上的表现飞速提升。这让一部分学者认为,大语言模型反映的语言习得与处理机制,相比起传统的生成语言学更接近语言规律的本质。但生成语言学和大语言模型在研究范式上存在根本性的不同。两种理论框架分别基于不同的研究对象和出发点,这决定了后续发展思路的差异,使得两者不可相互替代。大语言模型可以模拟人类语言运用的表现,却不能解释人类语言能力的本质。因此,大语言模型迅速发展的同时,继续推进生成语言学的研究仍有重要意义。The booming in the computing power and the database scale has been boosting the performance of modern large language models in natural language processing,leading some researchers to believe that the mechanism of large language models in processing natural language data is a better replacement of generative linguistics in pursuing the nature of language.But generative linguistics and large language models differ fundamentally in their research paradigms,which is reflected in their target of research and point of departure,hence two mutually irreplaceable subject branches.Large language models provide an accurate simulation of human behaviour in language use,but no explanation of human language capacity.Therefore,it is important to pursue the generative approach to linguistics together with the development of large language models.

关 键 词：大语言模型 生成语言学 递归 概率 

分 类 号：H08[语言文字—语言学] TP18[自动化与计算机技术—控制理论与控制工程]