Voz passiva analítica: análise léxico-funcional com implementação computacional no XLE

• Ano de defesa: 2019
• Autor(a) principal: Nogueira, Francisco Gleiberson dos Santos
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Voz passiva; Gramática Léxico-Funcional; XLE; Lexical Functional Grammar
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/49180

In this work, we propose a linguistic-computational analysis of the analytic passive voice in Brazilian Portuguese (hereinafter referred to as BP), whose status is discussed by Perini (2010), who stated that there is no passive voice in PB but rather a predicative adjective construction. Guided by the principles of Lexical Functional Grammar (LFG) and Computational Linguistics, we aim to verify the most suitable classification for the passive participle (PART-PASS) in passive periphrastic sentences. From a generative point of view, we defend the idea that PART-PASS belongs to the word class of adjectives and we present and discuss morphological and syntactic evidences (see ALENCAR, 2016) that corroborate this conjecture. In order to test the advantages of this type of analysis, we propose the implementation of two LFG grammars using the Xerox Linguistic Environment (XLE ) , which is the state of the art for the implementation of grammars in LFG (CROUCH et al., 2011): (a) a grammar in which PART-PASS is implemented as A (adjective), that is, the G-A grammar; and (b) a grammar in which PART-PASS is implemented as V (verb), the G-V grammar. Our basic hypothesis is that the G-A grammar presents a more efficient design. The development of both grammars was based on the adaptation of an existing LFG grammar for the French language (SCHWARZE & ALENCAR, 2016), through which the Portuguese phenomena were implemented incrementally. The testing of our grammars is done by analyzing two test sets, the positive and the negative, so that our parser must assign a satisfactory analysis to all the sentences contained in the first test set, but none in the second test set. The results show that the GA grammar presents a more efficient design when (see BUTT et al., 1999) some factors, such as CPU processing time and subtrees generated by the grammars, are taken into account.