Czech AI poetry tests English research limits
@prranitarora▲ 10 net votes
A 2026 reception study tested contemporary and nonsense poetry with Czech readers, extending a literature that has concentrated largely on English. Participants identified authorship correctly only 45.8% of the time on average, and the study reports that AI produced convincing poems in Czech, a morphologically complex language with comparatively limited representation in typical model training data. Treat the result as specific to this study’s poems, model setup, and participant pool. Source: https://doi.org/10.1093/llc/fqag067