Uma Investigação sobre Técnicas de Data Augmentation Aplicadas a Tradução Automática Português-LIBRAS

Abstract

The automatic translation from Portuguese to LIBRAS is extremely important for accessibility and inclusion of deaf individuals in society, but the scarcity of data and the high cost of building an authentic corpora pose significa challenges. Data Augmentation in Neural Machine Translation is the process of generating synthetic sentences to increase the quantity and diversity of the training set. This work investigates the use of data augmentation techniques to improve the performance of Portuguese-LIBRAS automatic translation using the BLEU metric. Among the techniques analyzed, back-translation and its combination with synonym substitution using part-of-speech tagging stood out as the most effective in enhancing the translation model and can be used increase the diversity of underrepresented datasets.