On the Development and Evaluation of a Brazilian Portuguese Discourse Parser

Abstract

Resumo: Apresentamos neste artigo o processo de desenvolvimento e avaliação de um analisador discursivo automático para o português brasileiro. Seguindo a Teoria de Estruturação Retórica, o DiZer é um sistema simbólico baseado na ocorrência de marcadores textuais, fazendo uso de templates discursivos extraídos de um corpus de textos científicos para identificar a construir a estrutura discursiva de textos. A avaliação do DiZer mostra resultados satisfatórios para textos científicos e jornalísticos, apesar do sistema não ter sido delineado para o gênero jornalístico, o que demonstra a portabilidade do sistema. Palavras-chave: análise discursiva automática, RSTAbstract: We present in this paper the development process and the evaluation procedure of a Brazilian Portuguese discourse parser called DiZer. Based on Rhetorical Structure Theory, DiZer is a symbolic cue phrase-based analyzer that makes use of discourse templates learned from a corpus of scientific texts to identify and build the discourse structure of texts. DiZer evaluation shows satisfactory results for scientific and news texts, even tough it was not designed for the latter, which demonstrates DiZer portability.

Annotation upon Annotation: Adding Signalling Information to a Corpus of Discourse Relations

“…The classification has a top-level breakdown into discourse markers, morphological, syntactic, semantic, lexical, genre and graphical features, plus heuristics specific to each relation. We started our annotation, as we explain in Section 5, by consulting previous studies for indication of what signalling devices have been found in corpora (Halliday & Hasan, 1976;Blakemore, 1987;Schiffrin, 1987;Fraser, 1990;Scott & de Souza, 1990;Dale, 1991;Blakemore, 1992;Sanders et al, 1992Sanders et al, , 1993Knott & Dale, 1994;Knott, 1996;Corston-Oliver, 1998a;Fraser, 1999;Marcu, 1999Marcu, , 2000bBateman et al, 2001;Schiffrin, 2001;Blakemore, 2002;Lapata & Lascarides, 2004;Polanyi et al, 2004;Sporleder & Lascarides, 2005;Fraser, 2006;Huong, 2007;Prasad et al, 2007;Pardo & Nunes, 2008;Sporleder & Lascarides, 2008;Teijssen et al, 2008;Fraser, 2009;Lin et al, 2009;Pitler et al, 2009;Louis et al, 2010;Prasad et al, 2010). Then, as we annotated more and more relations, we added to our classification.…”