CLASSIFICATION TREE FOR THE STUDY OF CORRELATION BETWEEN VARIABLES IN LINGUISTIC USAGE DATA: CONSTRIBUTIONS OF ORANGE DATA MINING SOFTWARE

CONTRIBUIÇÕES DO SOFTWARE ORANGE DATA MINING

Abstract

There is a basic prerogative from both the Sociolinguistic and Functionalist perspectives that phenomena undergoing variation and change do not occur randomly, but rather are motivated or conditioned by factors of a social, structural or cognitive nature. In this paper, we aim to show the potential contributions of the classification tree widget, available in the Orange Data Mining Software. The tool serves as a predictive method and, to this end, uses logistic regression calculations for the hierarchical description of how predictor variables potentially condition outcome variables. This statistical resource – also known by the term conditional inference regression tree (cf. Hothorn; Hornik; Zeileis, 2006; Speybroeck, 2012) – presents a user-friendly interface in Orange Data Mining, compared to the R language. As an illustration of the contribution of this tool to research in language usage, we describe the impact of three predictor variables on the instatiation of two linguistic constructions, both formed by the same sequence of elements – preposition sem + verb dicenci –, namely: the hypotactic adverbial negative modal or conditional clause (e.g.: ele saiu sem falar com ninguém); the discourse structuring marker of addition (e.g.: ele faltou hoje. Sem falar que, quando vem, sempre chega atrasado).