Análise automática de coerência textual em resumos científicos : avaliando quebras de linearidade

• Autor(a) principal: Silva, Leandro Lago da
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da Universidade Estadual de Maringá (RI-UEM)
• Texto Completo: http://repositorio.uem.br:8080/jspui/handle/1/2551
• topic: Coerência; Quebra de linearidade; Grade de entidades; Auxilio à escrita científica; SciPo; Brasil.; Coherence; Linearity break; Entity grid; Scientific writing support; SciPo; Brazil.; Ciências Exatas e da Terra; Ciência da Computação

Coherence makes a sequence of words, sentences or paragraphs, become a text, connecting the elements and giving meaning to the speech. To write coherent texts is a task that requires practice and skill in various linguistic aspects. One way to achieve these skills is to request aid for reviewers or for computational tools developed for this purpose. The Scientific Portuguese - SciPo is an example of writing tool for Portuguese that includes, among other features, a coherence analysis module (MAC) which detects potential problems of semantic coherence in scientific abstracts. Based on latent semantic analysis (LSA), MAC analyzes the semantic relationship between sentences of the abstract, in accordance with a predetermined set of dimensions. For one of the proposed dimensions for the MAC, called Linearity Break, the results obtained by LSA were unsatisfactory, suggesting the use of other coherence models. In this context, this project aimed at extending MAC by adding the Linearity Break dimension. The proposed approach for it combines the entity grid model with information from the abstract rhetorical structure, allowing MAC to generate suggestions pointing possible breaks linearityin specific regions of the abstract. Experimental results have shown that the proposed combination captures linearity breaks, and confirmed that the generated suggestions are useful, guiding users in writing texts with a higher level of coherence.