KAIST unveiled the next-generation braille translation engine K-Braille, which further advances braille translation technology for people with visual impairments.
On Mar. 13, a research team led by Professor Ga Hyeon-uk of the Rehabilitation Artificial Intelligence Lab in the School of Converged Human Resource Studies at KAIST said it developed K-Braille, which upgrades the technology that converts regular text (print) into braille.
The newly developed K-Braille goes beyond simple character substitution, converting to braille after understanding the meaning by grasping the structure and context of a sentence. While existing translation programs stop at one-to-one mapping of letters or symbols, K-Braille improves accuracy by analyzing entire sentences.
According to the research team, the engine can more precisely handle various exceptional cases that commonly cause errors in existing braille translation, such as sentences that mix foreign languages and Korean, complex symbol combinations, and notations for various units.
When the team extracted 17,943 sentences to verify K-Braille's practical compliance rate with braille translation rules, it recorded 100.0% accuracy. The translation morpheme structure similarity, which indicates how closely the structure of a braille sentence resembles the correct answer, averaged 99.81%.
Professor Ga is a researcher with a congenital, severe visual impairment who has felt more than anyone the meaning and potential of braille. Ga said, "Braille is not just a set of symbols for people with visual impairments, but a language for reading the world," adding, "Building on this result, we will advance the technology into a next-generation translation system that can handle not only mathematical formulas and scientific symbols but also musical scores."