THE DISAMBIGUATION OF HOMONYMOUS WORDS IN SENTENCES BY BRAZILIAN PORTUGUESE-LIBRAS AUTOMATIC TRANSLATION APPLICATIONS

ABSTRACT This study analyzes the Automatic Translation (AT) of homonymous words isolated and inserted in sentences, produced by Hand Talk and ProDeaf Mobile applications, both automatic translators from Brazilian Portuguese to Brazilian Sign Language (Libras), once previous studies have pointed to the absence of disambiguation strategies, use of fingerspelling and translation errors. This is an exploratory qualitative research based on studies of the lexicon of Libras, Automatic Translation, Natural Language Processing and Disambiguation, whose purpose is to deepen initial findings, under the same theoretical bias, in order to propose improvements on the AT of Portuguese Language-Libras. Thus, this study analyzed the AT of 38 homonymous words isolated and contextualized in 38 sentence pairs (two distinct meanings in each sentence pair). The results revealed that in AT of isolated words the applications generate the translation of only one of the meanings of the homonymous words (HT: 89% and PDM: 63%). In sentences, the percentage of correct word meanings, in both applications, is lower (HT: 82% and PDM: 60%) than in AT per isolated word and even lower (HT: 13% and PDM: 11 %) in the disambiguation of homonymous words. The findings of this research indicate the need to improve the Natural Language Processing system of the applications for the disambiguation of homonymous words. It is inferred that a more adequate AT at the level of isolated word and sentence can offer users more analogous translations to Libras, as the natural and official language of the deaf community in Brazil.