The widespread use of artificial intelligence (AI) systems in the production of scientific texts has attracted the attention of researchers to examine the linguistic characteristics of these texts in comparison with human writing. Yet, there are linguistic features that have been overlooked despite their significance for writing quality. This study, therefore, examines the syntactic complexity in the discussion section of research articles written by human authors and texts produced by AI, and analyzes the role of the type of author and scientific discipline in the formation of these patterns. For this purpose, a corpus of 160 discussion sections of research articles was compiled, including 40 texts from applied linguistics and 40 from biomedical engineering, published between 2019 and 2021 in reputable journals. For each human-written article in both disciplines, an AI-generated counterpart was created by feeding the results section of the same paper to an AI system through a standardized prompt, yielding 40 AI-produced discussion sections in applied linguistics and 40 in biomedical engineering. Syntactic complexity indices at the clausal and phrasal levels were then extracted from this parallel corpus based on the framework proposed by Biber and Gray (2016). The data were analyzed using descriptive statistics and a Uni-variate Generalized Linear Model (GLM). The results showed that human texts generally have higher clause complexity, while texts produced by AI tend to have higher phrase complexity and structural compactness. Moreover, author type has a significant effect on both types of syntactic complexity, while the effect of scientific discipline is only significant in clause complexity. The interaction effect of author type and scientific discipline is not observed at any levels of complexity. The findings of this study indicate that analyzing the differences between human writing and AI productions requires simultaneous attention to the language level, scientific genre, and disciplinary context, and can provide a basis for future research in the field of academic discourse analysis and evaluation of AI-produced writing.
