Aplicação de reconhecimento de padrões em um experimento linguístico

Abstract

The "Event-Related Potentials" (ERP) technique consists of the measurement of electrical biological signals obtained by electroencephalography (EEG), which are direct results of stimuli to sensory, cognitive or motor events. In this way, the ERP technique allows the non-invasive analysis of brain functioning. Based on the results obtained with Soto (2014), the treatment of these data and the extraction of ERP parameters, using the EEGLAB® and ERPLAB® tools based on the simulation program Matlab® (“matrix laboratory”). The result of the research was the obtaining of supervised and unsupervised classification scenarios for the classes proposed in the mentioned experiment and the comparative study and discussion of the classification results found, using the methodology proposed by Webb (2002). This work proved to be innovative in the area of Linguistics, since, at least until now, no similar work has been found in research databases such as IEEExplorer, Web of Science, Elesevier and Spring. Excellent results for supervised classification, with the "Random Forest" classifier reaching a total accuracy of 100%, followed by the "Multiclass Support Vector Machine" (MSVM) and "Naïve Bayes" classifiers, with both methods reaching precisions rreater than 96% were obtained. The results indicate that non-linear approaches were more adequate to classify Soto (2014) ERP configuration data and the results also open the possibility of analyzing signals from individuals with this ERP methodology associated to Pattern Recognition, with the possible application of this type of analysis in diagnostic tools, assessment of language learning, among others.