Kharms vs. neurokharms: neural network as a narrative laboratory

The article raises the question of the possibility of using narratives created by the transformer neural network “Porfiryevich” based on original author’s texts as a tool for studying the general patterns of text generation. The results of an experiment aimed at identifying the principles according to which a neural network imitates an act of artistic creativity are analyzed; It is established that the algorithms for generating narratives used by the neural network enter into significant correlations with the stylistic parameters and narrative strategies of artistic discourse. The study of text variants created by a neural network in continuation of D. Kharms’ miniatures from the series “Cases” allows us to trace how changing the parameters of the stimulus text influences the neural network’s choice of different narrative strategies - from the simplest grammatical imitation of the stimulus text to its semantic completion based on indexical signs of the original narrative. The results of the experiment reveal the potential logic of “neuronarrative”, and also define an open set of potential “unaccounted for possibilities” of narrative technologies, of which the act of artistic creation chooses only one. The results of text generation, which are demonstrated by a neural network trained on literary texts, can be considered as a kind of model of the creative process, representing the most pure writing technologies, free from author’s valences.