PrepNet.Br: uma proposta de representação semântica para as preposições do português

• Ano de defesa: 2018
• Autor(a) principal: Garcia, Débora Domiciano
• Orientador(a): Di Felippo, Ariani
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos; Câmpus São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Linguística - PPGL
• Palavras-chave em Português: Preposições; Rede semântica; Linguística cognitiva; Ontologia; Prepositions
• Palavras-chave em Inglês: Semantic network; Cognitive linguistics; Ontology
• Área do conhecimento CNPq: LINGUISTICA, LETRAS E ARTES::LINGUISTICA::LINGUISTICA APLICADA; LINGUISTICA, LETRAS E ARTES::LINGUISTICA::TEORIA E ANALISE LINGUISTICA
• Link de acesso: https://repositorio.ufscar.br/handle/ufscar/10006

In the last decades, the study of prepositions has been more modest when compared to the studies of nouns, verbs and adjectives, both in Linguistics and Natural Language Processing (NLP). However, some recent studies have shown that the semantic of prepositions is essential for the understanding of sentences and this can benefit computational applications. This thesis proposes the description of prepositions in terms of a semantic network inspired by WordNet (Princeton's Lexical Database for English, WN.Pr). In this proposal, it is argued that prepositions with similar spatial characteristics are synonymous and occupy the center of the network, from which their metaphorical extensions derive. In representational terms, the synonymy was modeled by means of synsets (as in WN.Pr), simple definitions (glosses) were assigned to each direction, and antonyms, hyperonymy, and hyponymy relations were established. Therefore, this research was developed in two research domains: Linguistic and Computational-Linguistics. In the first domain, (i) we study the theoretical cognitive knowledge that supports the spatial description of the basic sense of these items, and (ii) we identified their metaphorical senses based on a systematization of the different meanings attributed by Portuguese grammars. In the second domain, we proposed the systematization of the linguistic data in a conceptual hierarchy organized in a computationally treatable format implemented in Protégé, one of the most used editors for the creation and edition of ontologies. In total, 502 different senses were identified and indexed, creating the first version of PrepNet.Br, published online. Of these, 152 spatial meanings were attributed to 15 simple prepositions and 110 complex prepositions, while 350 are metaphorical meanings grouped into 114 classes. In addition, a total of 2358 sample sentences were selected. With the creation of this sense inventory, a new lexical resource was created for Brazilian Portuguese and the description of prepositions contributed to enrich the studies in Descriptive Linguistics. With the creation of this sense inventory, a new lexical resource was created for Brazilian Portuguese that may contribute to enrich the studies of Descriptive Linguistics and, even modestly, to the advance of NLP in Brazil.