O USO DE REDES NEURAIS AUTO- ORGANIZÁVEIS NA VISUALIZAÇÃO DE FORMAÇÃO DE AGRUPAMENTOS A PARTIR DO CONHECIMENTO ACENTUAL DE APRENDIZES BRASILEIROS DE INGLÊS

Abstract

! is study aimed at investigating how Brazilian learners ofEnglish organize their knowledge about lexical stress of a speci" c wordcategory at an early stage of L2 acquisition with the help of an unsuper-vised neural network, a self-organizing map (SOM), also called Kohonennetwork. ! e basic hypothesis tested was whether the parameterization ofthe speech signal from learner’s utterances through processing techniquessuch as Linear Predictive Coding (LPC), which consisted of the input ofthe network, would be e# ective in the classi" cation of learners and theirutterances. ! e study consisted of an empirical part and a computationalone. ! e participants were beginner students aged between 18 and 25.Preliminary results indicate that the combination of LPC+SOM allowedthe creation of well-de" ned category clusters, which is an important stepin data classi" cation to aid language pro" ciency level determination, andcomputer-assisted pronunciation teaching.