RHeSumaRST: um sumarizador automático de estruturas RST

• Ano de defesa: 2005
• Autor(a) principal: Seno, Eloize Rossi Marques
• Orientador(a): Rino, Lúcia Helena Machado
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de São Carlos
• Programa de Pós-Graduação: Programa de Pós-Graduação em Ciência da Computação - PPGCC
• País: BR
• Palavras-chave em Português: Processamento da linguagem natural (Computação); Sumarização automática; Inteligência artificial
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: https://repositorio.ufscar.br/handle/20.500.14289/623

This work presents an automatic summarization model based on both the Rhetorical Structure Theory RST and the Veins Theory VT. RST allows inter-relating discourse units by means of rhetorical relations. These, in turn, mirror meaning relations between those units. Adding to RST, VT delimits the domain of referential accessibility of each discourse unit of an RST tree, resulting in its vein . A vein signals, thus, the limits of a discourse unit that may enclose its anaphora antecedents. The automatic summarization model also embeds Marcu s model of salience: once a discourse is structured as an RST tree, its units are classified according to their salience by considering the its deep in the tree. The model consists of a set of pruning heuristics that aim at identifying superfluous information in an RST tree of a text. In excluding them, the resulting summary RST tree and, thus, the text summary, should preserve the co-referential chains. In this way, the heuristics are driven by both, specific features of RST relations and constraints on the referential accessibility provided by VT. The main contributions of this work include the proposal of the AS model itself and the availability of a prototype for its automatic exploration.