Tradutores automáticos de libras baseados em inteligência artificial - impactos na inclusão digital

Abstract

The development of automated translation applications for Brazilian Sign Language (Libras) has impacted the social inclusion of the deaf community in digital environments. Although tools such as Hand Talk and VLibras are designed to promote accessibility and autonomy, research with deaf users shows widespread dissatisfaction, with 75% of respondents expressing discontent with translation quality. The main criticism is that current applications generate so-called “Signed Portuguese,” a word-for-word translation that does not conform to the visual-spatial grammatical structure of Libras, which usually follows an SVO (Subject-Verb-Object) or SOV (Subject-Object-Verb) order. Technical limitations go beyond syntax, including the inability to use facial and body expressions—essential for prosody and meaning—and the inability to reproduce classifiers, a complex linguistic feature of Libras. Given these shortcomings, the community views these applications mainly as dictionaries for isolated gestures or tools for beginners, agreeing that, for now, technology cannot replace the complexity and subjectivity of human translation. To overcome these obstacles, new research focuses on more sophisticated applications of Artificial Intelligence (AI). Recent projects propose developing minimum viable products (MVPs) that use AI to first restructure Portuguese sentences into the correct Libras grammatical structure and then translate them into gestures. Tests using this approach have shown comprehension rates above 90% for fluent users. Continued studies, combined with advances in facial recognition technology that can enhance avatar expressions, are expected to promote more fluent and authentic communication, aiming for digital inclusion that respects the linguistic and cultural integrity of the deaf community.