Identificação computacional de padrões interníveis em textos da literatura brasileira

Abstract

In literary texts, in poetry or prose, there is an intense presence of poetic devices and recurring linguistic resources at different linguistic levels. The identification of these linguistic resources with the help of computational tools can point out, through quantifiable analyses, patterns of relationships between these levels. The objective of this dissertation is to propose a computational method that allows the identification and correlation of textual patterns between linguistic levels in Brazilian literature texts. To this end, textual characteristics were extracted at different linguistic levels based on the quantification of occurrences through absolute and relative frequencies, both for the full text and for text excerpts, followed by correlation analysis of these quantified characteristics to identify patterns of parallelism between them. The results obtained in order to demonstrate this computational method were extracted from three Brazilian literary works (Dom Casmurro, Macuna´ıma and Os Sert˜oes). These results contribute to the understanding of the various facets of the method, highlighting its ability to identify and correlate patterns at multiple linguistic levels, while demonstrating the variability of possible results, allowing a quantitative analysis of the patterns present. This research has the potential to open paths for studies in textual analysis, introducing a quantitative approach into a predominantly qualitative field.