Anticipatory and theme-specific neural oscillations predict aesthetic evaluation of poetry

Poetry condenses language into minimal forms, evoking emotion, imagery, and aesthetic judgment, yet the neural basis of such evaluations remains poorly understood. We investigated how the brain evaluates two structurally matched but thematically distinct poetic forms: nature-themed Haiku and emotion-themed Senryu. Participants read poems and rated them across five dimensions-aesthetic appeal, vivid imagery, being moved, originality, and creativity-while EEG was recorded. Using multiclass gradient-boosted tree models with SHapley Additive exPlanations, we predicted evaluative ratings from oscillatory neural features across temporal windows and scalp regions. Models outperformed linear baselines and showed limited cross-theme generalization, indicating content-specific neural encoding. Distinct processing patterns emerged: Senryu showed stronger beta-band contributions, whereas Haiku engaged more distributed multifrequency dynamics. Temporal profiles also differed, with Haiku showing sustained engagement across reading and contemplation phases and Senryu showing earlier evaluative resolution during reading. Prestimulus neural activity contributed to prediction of subsequent evaluations, suggesting a role for anticipatory brain states in aesthetic evaluation. Across poems, evaluative dimensions converged on a dominant shared axis that was reliably predicted from neural features. Together, these findings suggest that aesthetic evaluation of poetry reflects an interaction between anticipatory neural states, content-specific oscillatory dynamics, and dimension-specific processes organized around a shared evaluative axis. This work establishes poetry as a tractable model system for studying how the brain constructs meaning and value from minimal linguistic input.