Computational modeling of 19th-century nheengatu from the lower amazon region using universal dependencies

Abstract

This thesis aims to analyze the morphosyntax of Nheengatu using the Universal Dependencies

(UD) framework. The proposal is situated within the field of Digital Humanities and seeks to con- tribute to the linguistic description of this Indigenous language, as well as to the development of

computational resources for its morphosyntactic annotation. Given that Nheengatu is undergoing a reduction in the number of speakers and in intergenerational transmission (NAVARRO, 2012; EBERHARD et al., 2025), we claim that the development of natural language processing tools can contribute to its documentation and digital inclusion, particularly through the application of computational formalisms that are increasingly adopted in the state of the art, such as the UD framework. From a theoretical perspective, this study is grounded in the Universal Dependencies theory (NIVRE et al., 2017a; NIVRE et al., 2017b) and in previous linguistic descriptions, especially those by Cruz (2011), Stradelli (2014), Navarro (2016), and Avila (2021). The adopted corpus consists of 19th-century records of Nheengatu spoken in the region of the lower Amazon River, as documented in Hartt (1938). The methodology involves the orthographic normalization of these sentences based on Avila (2021), followed by morphosyntactic annotation and revision according to the principles of the UD project. The practical objective is to expand the existing Nheengatu treebank in the UD collection (ALENCAR, 2023; ALENCAR, 2024a), as well as to identify and describe syntactic patterns characteristic of this historical variety of the language. In doing so, this study aims to promote the inclusion of Nheengatu in the current landscape of computational-linguistic research, building upon the already attested applicability of the UD framework to the morphosyntactic analysis of minority languages (GERARDI et al., 2022; RODRÍGUEZ et al., 2022; SANTOS et al., 2024).