Descrição, classificação e processamento automático das construções com o verbo dar em Português Brasileiro

Abstract

This PhD thesis starts from the observation of constructions with the verb dar in Brazilian Portuguese and it proposes an analysis and classification of these constructions. The analysis of all constructions found in corpus is systematic and consistent, based on Transformational Grammar (HARRIS, 1964) and on Lexicon-Grammar approach (GROSS, 1975; GROSS, 1981). Both theories adopt formal and experimental criteria to make reproducible the identification of base sentences and the classification of constructions with the verb dar. The classification was based on structural and syntactic properties, leading to four classes or categories: (i) full verb or distributional verb, from which 8 verbal senses could be identified; (ii) causative operator verb, whose constructions could be subclassified in 4 groups, depending on the semantic type of the predicative noun; (iii) the verb as a fixed element, in idioms or in proverbs; and (iv) support verb, which is selected by a predicative noun. The first part of this thesis analyzes the constructions with the verb dar as a full verb, as a causative operator verb and as a fixed verb. The second part of this thesis debribes the support verb constructions (SVC), which are formed by the support verb dar (Vsup) and the predicative noun (Npred). This description is extensive, considering all the occurrences in the corpus adopted and identifying the main formal, distributional and transformational SVC properties. These properties were described and formalized in a Lexicon-Grammar table, which is a binary matrix: each line corresponds to a lexical entry (Npred) and each column corresponds to a syntactic property of the construction. In the intersection between each line and each column, we sign “+” or “-”, respectively, if the property is verified or not. The third part of the thesis presents an automatic syntactic analysis of SVC, by using an approach based in dependency rules between its constituents. The rules are automatically generated from the L-G matrix data and, then, those rules are used by the parser in order to extract the dependency between the Npred and the Vsup. Thus, this work aims to advance the state of the art of the general verbal classification in Portuguese and Automatic Processing of Natural Languages, to contribute to the development of computational-lexical resources (the Lexicon-Grammar of nominal constructions) and with automatic parsing.