LLM Reconstructions of the Text Narrative With and Without the Time Markers

Abstract

Annotation Traditional narrative structure is often based on a linear time sequence of events. However, there are many books in which the time axis is replaced by a system of causeand-effect and associative relationships between events. Examples of this approach include works such as Cloud Atlas (Cloud Atlas) by David Mitchell, The Torah and The Master and Margarita by Mikhail Bulgakov or the Hebrew Tora. This article presents a model for text reconstruction based on a network structure, in which cause-and-effect and associative relationships act as the main coordinates. The presented model uses large language models (LLMs). The methodology involves the creation of a semantic network using LLM and subsequent text reconstruction, which is illustrated by the example of the reconstruction of a short story consisting of two plots. We consider how plots can be modeled through such networks without regard to traditional chronological time. The article presents a mathematical model that describes the process of reconstruction and the paradoxes of time that arise as a result of the mixing of associative and causal connections, which leads to the creation of new spaces of meaning. This model allows us to reconstruct a complete narrative without the need to take into account chronological time, which opens up new opportunities for analysis, reconstruction, modification and interpretation of texts.

Keywords: Text reconstruction, hypergraphs, cause-and-effect relationships, associative connections, time paradoxes, nonlinear storytelling, world reconstruction, LLM