Semantic-related challenges in computational intelligence: a transdisciplinary approach

• Ano de defesa: 2023
• Autor(a) principal: Silveira, Tiago Buatim Nion da
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Universidade Tecnológica Federal do Paraná; Curitiba; Brasil; Programa de Pós-Graduação em Engenharia Elétrica e Informática Industrial; UTFPR
• Palavras-chave em Português: Computação semântica; Processamento de linguagem natural (Computação); Semântica; Visão por computador; Inteligência; Inteligência computacional; Inteligência artificial; Abordagem interdisciplinar do conhecimento; Semantic computing; Natural language processing (Computer science); Semantics; Computer vision; Intellect; Computational intelligence; Artificial intelligence; Interdisciplinary approach to knowledge; CNPQ::CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO; Engenharia Elétrica
• Link de acesso: http://repositorio.utfpr.edu.br/jspui/handle/1/33531

In this thesis, we identify and address semantic challenges in computational intelligence, especially those in natural language processing and computer vision, known as the semantic gap. Starting from the hypothesis that semantic attribution is intelligent behavior and that the term intelligence is used from different perspectives in psychology and computing, we carried out in-depth research through which we identified that there is no consensus for intelligence, even in the same area. To meet this need, we proposed a comprehensive definition of intelligence, formalized as the ability of an autonomous agent to process external and internal information to obtain optimal environmental adaptation. This definition applies to any agent, for which we propose categorization into human, artificial, and biological intelligent agents. In our review, the CHC model of intelligence is cited as the gold standard in intelligence studies and psychological assessment. However, its application in the computational area is challenging. Furthermore, we categorized its 83 factors into five aspects of intelligence: formal, semantic, contextual, social or affective, and processing resources. Our focus then becomes only semantic intelligence, which we approach from a transdisciplinary perspective. The challenge of such an approach is that different areas have different epistemic bases, which we overcome by proposing paradigmatic analysis: the identification of studies and theories in three groups of paradigms related to semantics: ontological, about the nature of meaning and which can be objective or subjective; of measurement, on the observation of meaning, which can be deterministic or inferential; and computational, about the decision or attribution of meaning, which can be formal-logical or based on information processing. We comprehensively explore and categorize semantics-related theories from linguistics and psychology using this methodology. The organization of knowledge between these fields is one of the main contributions of this thesis, bringing new perspectives to semantic research. Consequently, we found that not all paradigms can represent semantic content other than partially. In this way, we proposed a formal model of semantic attribution, in which the semantic content is decomposed into its objective, subjective, and contextual components. The proposed model is verified through a factor analysis of the Glasgow Norms – a corpus that relates words to different psycholinguistic variables. Using these factors, we proposed a semantic content analysis and classification. Finally, we offer a reinterpretation of the semantic gap, often found in computational intelligence tasks. Such reinterpretation was carried out through the paradigmatic analysis of the studies researched and the analysis of the ontological parameters that influence semantics through pareidolia. The thesis concludes by proposing a new definition for the semantic gap sustained in the formal model developed for semantics.