Corpus Linguistics and Artificial Intelligence

Abstract This article argues that a register-based Multi-Dimensional (MD) description is a suitable route for characterizing AI-generated language in corpus linguistics. The argument is illustrated with two sample studies: a grammar-oriented investigation that applies traditional MD analysis to English-as-a-foreign-language textbook texts and a discourse-oriented analysis that relies on lexical MD analysis to explore AI-generated pop music lyrics. In both cases, the results reveal sharp differences between AI-generated and human language. In the EFL texts, AI-written texts are more informational, abstract, and impersonal whereas human texts display interpersonal awareness, stance, and engagement. In the pop lyrics, AI generates moralized empowerment discourses that recast historical conflicts depicted in rap music as generalized virtue narratives. In both cases, AI demonstrates signs of register deficit (a limited awareness of register variation due to shallow knowledge of the linguistic constituency of human registers) and register metamorphosis (generation of texts that resemble one register on the surface but are realized linguistically as another).