Identifying Encapsulating Anaphoras: comparison between automated and human analysis

Abstract

Encapsulating anaphoras are the focus of this study, highlighting their relevance for the analysis and teaching of writing. Considering the advances in the use of Artificial Intelligence (AI) in computational linguistics, the research is justified by the need to integrate innovative methodologies in textual analysis. The motivation arises from the gap in the application of large language models (LLMs) to identify and categorize descriptive and opinionated encapsulations in essays. The study proposes a hybrid approach that combines human analysis, rich in contextual nuances, with the capabilities and scalability of LLMs using zero-shot and few-shot prompts. The experiments carried out with Enem essays show that the use of few-shot prompts significantly improves the identification of encapsulating anaphoras by LLMs, when compared to the use of zero-shot prompts, but still below that observed in human analyses. In short, this work seeks to contribute to the advancement of research in textual linguistics and AI, offering a new perspective for the analysis of written texts and demonstrating the potential of combining human and computational methods to identify complex linguistic patterns.

Keywords: Encapsulated Anaphoras; Textual Analysis; Enem Essay; Large Language Model.