Modelagem e Desenvolvimento de uma Base de Dados para Tradução Automática de Conteúdos Digitais para Língua Brasileira de Sinais

Abstract

This research presents a proposal for modeling and developing a bilingual textual database to support the development of automatic translators from Portuguese to Libras, using a textual notation in Libras, called transcription system in gloss. For this, initially some rules were defined to specify grammatical aspects that are important and characteristic of Libras in this notation. Based on this notation, a bilingual textual database was developed using a team of 10 Libras interpreters, resulting in a bilingual corpus containing 57,206 examples of translations in Portuguese and their corresponding gloss representation in Libras. This database was then used and applied for training a new VLibras automatic translator model, and the impacts produced in VLibras in relation to the treatment of these grammatical aspects of Libras were evaluated with deaf users. The results indicated that users rated that they understood the translated sentences better when the database was incorporated into VLibras.