Identificação automática de construções de estrutura argumental: um experimento a partir da modelagem linguístico-computacional das construções transitiva direta ativa, ergativa e de argumento cindido

Abstract

This work is included in the theoretical and methodological discussions of FrameNet Brasil (FN-Br) and the Constructicon of Brazilian Portuguese. It aims to (i) provide a linguistic-computational modeling necessary for the treatment of the Active Transitive, Ergative and Split Object Constructions (CAC, in Portuguese) in the Constructicon of Brazilian Portuguese; (ii) present and discuss the results of a discovery-task of constructions by machine whose purpose is to identify instances of the constructions mentioned above; (iii) discuss the implications of this task for the development of constructicons. Thus, this dissertation is based on the constructional approach of Berkeley Construction Grammar proposed by Fillmore, Kay and O’Connor (1988), by Kay and Fillmore (1999) and Fillmore (2013). Our proposal is justified by the lack of studies that integrate linguistic descriptions with models of computational representation cognitively plausible, enabling the treatment of issues that surpass the representation models centered exclusively on syntactic aspects. Hence, this works falls within the framework of Computational Linguistics which, according Dias da Silva (1996), involves the integration of issues in three domains: Linguistic, Representational and Implementational. This research is inserted in the representational stage, once the linguistic stage has already been proposed by Sampaio (2010), Ferreira (2009), Castilho (2010) and Perini (2010). Given this framework, our objective is to computationally model the Active Transitive, Ergative and CAC into the constructional base of FrameNet, so that, later, in the Implementational stage – not developed in this dissertation, but whose results are discussed here – could be used in a constructional discovery-task by machine through the parser under development by FN-Br, the CARMA.