O uso de redes neurais auto-organizáveis na visualização de formação de agrupamentos a partir do conhecimento acentual de aprendizes brasileiros de Inglês

Abstract

This study aimed at investigating how Brazilian learners of English organize their knowledge about lexical stress of a specific word category at an early stage of L2 acquisition with the help of an unsuper- vised neural network, a self-organizing map (SOM), also called Kohonen network. The basic hypothesis tested was whether the parameterization of the speech signal from learner’s utterances through processing techniques such as Linear Predictive Coding (LPC), which consisted of the input of the network, would be effective in the classification of learners and their utterances. The study consisted of an empirical part and a computational one. The participants were beginner students aged between 18 and 25. Preliminary results indicate that the combination of LPC+SOM allowed the creation of well-defined category clusters, which is an important step in data classification to aid language proficiency level determination, and computer-assisted pronunciation teaching.