Assessment of the Treatment of Grammatical Aspects of Machine Translators to Libras

Currently, a set of technologies has been developed with the aim of reducing barriers to access to information for deaf people, such as machine tools for sign languages. However, these technologies have some limitations related to the difficult of handling some specific grammatical aspects of the sign languages, which can make the translations less fluent, and influence the deaf users experience. To address this problem, this study analyzes the machine translation of contents from Brazilian Portuguese (Pt-br) into Brazilian Sign Language (Libras) performed by three machine translators: ProDeaf, HandTalk and VLibras. More specifically, we performed an experiment with some Brazilian human interpreters that evaluate the treatment of some specific grammatical aspects in these three applications. As a result, we observed a significant weakness in the evaluation regarding the adequacy treatment of homonymous words, denial adverbs and directional verbs in the translations performed by the applications, which indicates the need for these tools to improve in the treatment of these grammatical aspects.