Linguistic description and machine learning: analysis of spanish locative verbs

• Autor(a) principal: Rodrigues, Roana
• Data de Publicação: 2022
• Outros Autores: Souza, Jackson Wilke da Cruz, Santos, Roney Lira de Sales
• Tipo de documento: Artigo
• Idioma: por
• Título da fonte: Cadernos de Estudos Linguísticos
• Texto Completo: https://periodicos.sbu.unicamp.br/ojs/index.php/cel/article/view/8666995
• topic: Machine learning; Syntax; Lexicon-grammar; Aprendizaje de máquina; Sintaxis; Léxico-gramática; Aprendizado de máquina; Sintaxe; Léxico-gramática

In order to explain the relations established between descriptive linguistics and machine learning, this article presents the results of a research that analyzes an algorithm generated based on the human classification of locative verbal constructions of the Spanish language. The data used in the investigation were taken from Rodrigues (2019), which presents a manual analysis and description of 318 instances that are constituted by verbs that necessarily select an argument interpreted as place (poner, salir, entrar, cage, etc.), distributed in 10 distinct classes, according to their structural, distributional and transformational properties. Based on the symbolic paradigm and using weka software, the data allowed the generation of two proposed rules of the JRip algorithm: without and with the selection of attributes. Both procedures generated 10 composite rules and evaluated the measurements of precision, coverage, f-measure and confusion matrix of the algorithms created. The algorithm without the selection of attributes presented 100% accuracy, demonstrating that the linguistic data present a coherent description and classification. The algorithm with the selection of attributes, having 96.54% accuracy, made it possible, in addition to exposing the most relevant linguistic properties for classification purposes, to analyze the most sensitive cases for class distinction, culminating in the survey of six descriptive aspects of review and/or refinement of data that should be analyzed in future linguistic studies. Thus, this investigation helped, more specifically, in the improvement of the description of the locative verbal constructions of the Spanish language and made it possible to demonstrate that the relationship between human description and machine learning is not only about the importance of description as an insum for the machine, but mainly on how it is possible to use algorithms (and their evaluation measures) to validate and improve the description of different phenomena of natural languages.