AI generated texts are being used more for academic and professional communication, but it remains an open question how machine learning produces linguistic structure from how natural language is processed. Large language models produce fluent and syntactically sensible texts, but to what degree their intervention in morphology is skillfully linguistic, rather than statistical pattern, based remains to be determined. This work studies the morphological touch on the artificial intelligence-generated texts vs human written texts. This is a comparative corpus-based study based on a small, controlled corpus representing comparable registers and adopting a quantitative design. It currently analyzes derivational morphology, inflectional morphology, nominalization, irregular formations, morphological productivity, and overgeneralization. These results suggest that this class of texts are highly conservative, and produce regular, frequent, surface-level morphological patterns. While basic inflectional forms like number, tense, and agreement are usually well formed in the AI corpus, this regularity should not be taken as evidence of greater linguistic accuracy. In fact, it is probabilistically choosing among patterns seen in human language data and in relatively human-designed prompts. Disparate sections of human-written texts, however, exhibit greater context-sensitive morphological variation, more controlled use of irregular forms, and stronger associations between morphological choice and intention, register and communicative purpose. This concludes that morphology is a useful domain for separating human linguistic competence from machine surface regularity.
