DOI: https://doi.org/10.36347/sjahss.2026.v14i02.001

Pages: 15-34

As AI-mediated writing becomes increasingly visible in literary and digital reading contexts, it is critical to understand how readers experience AI-involved narratives across languages. This study examined whether narrative prediction, emotional engagement, and perceived narrative voice differ between human-authored and AI-mediated short stories in two language cohorts: English and Mandarin Chinese. We constructed 12 tightly matched story pairs (6 per language), controlling for length, sentence count, readability, and baseline lexical properties. A large online sample was recruited (N = 652), with exclusions applied using pre-registered criteria, yielding a final analytic sample of N = 528 (English n = 264; Mandarin n = 264). Narrative prediction was assessed using a Cloze Probability Task. Across languages, AI-mediated texts showed lower cloze predictability than human-authored texts, with a significant main effect of Text Type (β = -0.076, p < .001) and a significant Text Type × Language interaction (β = -0.035, p = .009), reflecting a larger predictability penalty in English. Subjective outcomes showed robust main effects of Text Type for Narrative Engagement (β = -0.414, p < .001) and Emotional Intensity (β = -0.375, p < .001) without cross-linguistic interaction, indicating a consistent experiential reduction across cohorts. Narrative voice exhibited the strongest AI-related penalties across Authenticity, Stylistic Naturalness, and Perspectival Coherence (all p < .001), with a language-sensitive interaction for coherence (β = -0.105, p = .018). Moderation analyses revealed that AI familiarity attenuated subjective penalties for engagement, emotion, and voice authenticity/naturalness, but did not significantly moderate cloze predictability. An integrative effect-size synthesis and the Narrative Triad Divergence Index further demonstrated a larger overall AI-related divergence in English (NTDI = 1.03) than Mandarin (NTDI = 0.79). Collectively, these findings sug