The Neural Poetry of Grammar and the Neural Grammar of Poetry

This article raises the question of the possibility of using poetic texts generated by neural networks as an analytical tool for verifying scholarly hypotheses and observations about the characteristics of the reception of a poetic text. The article presents the results of an experiment aimed at determining the ability of a neural network to imitate the "poetry of grammar," that is, to reproduce recognizable grammatical features of a source text. The authors conclude that the results of the work of a neural network with the "poetry of grammar" is more similar to the reception mechanisms displayed by an "average" reader than to the creative competence of a poet. The neural network is able to "read" the surface structures of a text, while the realm of the depth of poetic grammar remains closed off to it.