| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 688.37 KB | Adobe PDF |
Authors
Abstract(s)
Atualmente, a vasta quantidade de textos online à disposição de qualquer organização ou indivíduo
tornou-se um enorme desafio. O consumidor/leitor encontra-se num labirinto de informações
não estruturada (texto) em constante crescimento, muita dela de baixa qualidade. Em
vários domínios, o leitor enfrenta um desafio ainda maior, sempre que necessitar de selecionar
informações textuais confiáveis e de alta qualidade. É um processo trabalhoso, geralmente
atingindo uma eficácia limitada. A avaliação sistemática e a seleção de documentos de alta
qualidade tornaram-se uma tarefa impossível de ser executada manualmente por qualquer ser
humano.
Portanto, o objetivo principal deste trabalho foi explorar os marcadores linguísticos que permitem
uma caracterização efetiva da qualidade e estética contida no texto. Assim, apresentamos
aqui uma análise exploratória e comparativa de um conjunto de vinte e um marcadores para avaliar
a qualidade e a estética no texto. Também medimos o desempenho de auto-semelhança desses
marcadores, em corpora, através de estimadores eficientes do parâmetro de Hurst. Quanto
ao material experimental, reunimos e usamos dois corpora diferentes em termos de qualidade
de escrita. Um corpus com alto padrão de qualidade, contendo obras clássicas da literatura, incluindo
várias obras-primas do Prêmio Nobel. O outro com texto de qualidade inferior, vindo de
vários internautas, incluindo textos escritos em blogs e por autores mais jovens e inexperientes.
Os marcadores experimentados são agrupados em cinco tipos: vocabulário, coesão, coerência,
complexidade sintática e figura de linguagem. As medições forneceram resultados muito interessantes,
levando-nos a concluir que existem marcadores linguísticos de alta qualidade, muito
úteis para distinguir entre textos bons e maus. O uso desses marcadores permitirá a implementação
de sistemas capazes de realizar essa classificação, de forma automática e com alta
qualidade.
Nowadays the vast amount of online text at the disposal of any organization or individual has become a huge challenge. The consumer/reader finds itself in a maze of constantly growing unstructured information (text), much of it of poor quality. In several domains, the reader faces an even greater challenge, whenever he needs to select reliable and high-quality textual information. It is a laborious process, usually reaching a limited effectiveness. Systematically assessing and selecting high quality documents have become an impossible task to be performed manually by any human being. Therefore, the main goal of this work was to explore the linguistic markers that enable an effective characterization of the quality and aesthetics contained in text. Thus, we present here an exploratory and comparative analysis of a set of twenty-one markers for assessing the quality and aesthetics in text. We have also measured the Self-Similarity performance of these markers, in corpora, through efficient estimators of the Hurst parameter. As for the experimental material, we have assembled and used two different corpora in terms of writing quality. One corpus having high quality standards, containing classic works of literature, including several Nobel-Prize masterpieces. The other one with text of inferior quality, coming from multiple internauts, including text written in blogs and by younger and inexperienced authors. The experimented markers are grouped in five types: vocabulary, cohesion, coherence, syntactic complexity and figure of speech. Measurements provided very interesting results, leading us to conclude that there are high quality linguistic markers very useful for distinguishing between good and bad texts. The use of these markers will enable the implementation of systems capable of performing this classification, automatically and with high quality.
Nowadays the vast amount of online text at the disposal of any organization or individual has become a huge challenge. The consumer/reader finds itself in a maze of constantly growing unstructured information (text), much of it of poor quality. In several domains, the reader faces an even greater challenge, whenever he needs to select reliable and high-quality textual information. It is a laborious process, usually reaching a limited effectiveness. Systematically assessing and selecting high quality documents have become an impossible task to be performed manually by any human being. Therefore, the main goal of this work was to explore the linguistic markers that enable an effective characterization of the quality and aesthetics contained in text. Thus, we present here an exploratory and comparative analysis of a set of twenty-one markers for assessing the quality and aesthetics in text. We have also measured the Self-Similarity performance of these markers, in corpora, through efficient estimators of the Hurst parameter. As for the experimental material, we have assembled and used two different corpora in terms of writing quality. One corpus having high quality standards, containing classic works of literature, including several Nobel-Prize masterpieces. The other one with text of inferior quality, coming from multiple internauts, including text written in blogs and by younger and inexperienced authors. The experimented markers are grouped in five types: vocabulary, cohesion, coherence, syntactic complexity and figure of speech. Measurements provided very interesting results, leading us to conclude that there are high quality linguistic markers very useful for distinguishing between good and bad texts. The use of these markers will enable the implementation of systems capable of performing this classification, automatically and with high quality.
Description
Keywords
Auto-Semelhança e Figuras de Estilo Coerência Coesão Complexidade Sintática Qualidade do Texto Vocabulário