ChatGPT的“理解”与“意义”:论其生成语言背后的形式、功能与立场

From the Perspective of the Language in the Analysis of ChatGPT

ChatGPT是最近网络上的热门话题。本文从语言学的形式主义、功能主义和批评话语研究三个层次,探讨了ChatGPT在理解自然语言意义方面展现出的不足之处。首先,ChatGPT的语言机制是依赖于对英语表面的语言规则和统计规律的预训练,注重培养大型语言模型的“形式语言能力”,而非与人类相同的语言认知能力;其次,ChatGPT囿于语料库的信息检索,缺少语境的辨别能力和使用能力,因此无法实现语言的三大元功能。为此,本文依据系统功能语言学对语义的定义,构建了适用于计算机语言学的函数表达式,以期更好地实现自然语言理解和生成中的一些“功能语言能力”。此外,研究者也辩证地发现,ChatGPT作为一种基于网络大数据的人工智能技术,在其生成的文本背后,它的训练数据不可避免地受到了权力和意识形态的影响,导致其倾向于某种特定的政治立场。这需要用户提高意识形态的分辨能力。

ChatGPT has recently become a hot topic on the Internet.This paper discusses the shortcomings of ChatGPT in understanding the meaning of natural language from the perspectives of formalism,functionalism and critical discourse studies.First,ChatGPT relies mainly on pre-training in the formal language rules and statistical patterns of English,emphasizing the development of the “formal language competence” of large language models,rather than the language cognitive abilities as human beings.Meanwhile,limited by its corpus-based information retrieval,ChatGPT lacks the ability to discern and utilize context,thus failing to achieve the three metafunctions of language.In response,this article constructs function expressions applicable to computational linguistics based on the definition of semantics in Systemic Functional Linguistics,in order to better address “functional language competence” in natural language understanding and generation.Additionally,researchers have also dialectically discovered that ChatGPT,as an artificial intelligence technology based on large-scale Internet data,inevitably has its training data influenced by power and ideology,leading to specific political stances.This requires users to enhance their ability to discern ideological influences.