Generating dynamic and creative texts with Large Language Models (LLMs): a systematic review

Abstract

This article presents a systematic review of the literature on the application of Large Language Models (LLMs) in generating and identifying creative and dynamic texts based on symbolic, narrative, and literary imaginary references, considering publications from 2014 to 2024. The analysis encompasses two main stages: (1) automated bibliographic screening and curation with the support of tools based on artificial intelligence, and (2) critical analysis of the approaches, challenges and opportunities present in the selected literature. The tools used made it possible to manage references, extract metadata, and identify thematic patterns. The results indicate a predominance of studies focused on the creative generation of texts and the detection of figurative language, such as metaphors, with increasing use of hybrid models that combine symbolic and sub-symbolic structures. The conclusion is that, despite the potential of LLMs for computational creativity, challenges persist such as low interpretability and poor adaptation to non-hegemonic cultural contexts.

Keywords:
Large Language Models; Literary imaginary; Systematic review