体认语言学视阈下ChatGPT语言生成及性能研究

An Embodied Cognitive Linguistic Approach to ChatGPT Language Generation and Performance

近来兴起的ChatGPT改变了人们对语言生成的认知。本研究基于体认语言学的语言生成观审视ChatGPT的语言生成和语言性能,以期获得语言研究和人工智能融合的启示。本研究首先回溯了ChatGPT语言生成技术的三个环节,分析了相关的理论语言学内涵;其次,结合体认语言学从现实到认知再到语言的语言生成观,分析人们真实使用的语言是现实、认知的产物,进而结合ChatGPT语言生成的三个环节,分析其语言生成基本遵循从“语言到语言”的生成范式,其语言生成只有少量的认知间接参与。ChatGPT庞大的语料使其具备强大的语言生成能力,但是由于缺乏现实和认知作为基础,会形成与事实不相符、语义混淆、无意义废话等性能偏误。最后,本研究认为优化ChatGPT语言性能的关键要解决其具身认知难题。

The recent ChatGPT has changed our view of language generation.The present study reexamines the language generation and performance of ChatGPT based on the language generation theory of ECL(embodied cognitive linguistics),for the purpose of furthering the integration of linguistic research and artificial intelligence.Firstly,the present study reviews the three phases of the language generation technology of ChatGPT and its linguistic implications.Secondly,the present study holds that the daily language is the result of reality and cognition based on the language generation theory of ECL,specifically from reality to cognition and then to language.Thirdly,by examining the three phases of ChatGPT's language generation,the present study deems that its language generation follows a"language to language"pattern with litle indirect involvement of cognition.Thus,though ChatGPT is empowered with great language generation ability,performance errors are bound to appear,such as falsity,confusion as well as nonsense.Finally,the present study considers the key to overcoming these performance errors is for ChatGPT to realize embodied cognition.