A research team led by Hyunwook Kah (가현욱), a professor at KAIST’s School of Transdisciplinary Studies and the Rehabilitation AI Laboratory, developed the next-generation braille translation engine K-Braille (K-Braille) and completed large-scale performance verification.

KAIST said on March 13 that a research team led by Hyunwook Kah (가현욱), a professor at the School of Transdisciplinary Studies and the Rehabilitation Artificial Intelligence (AI) Laboratory, developed K-Braille (K-Braille), a next-generation braille translation engine that advances braille transcription technology, and completed large-scale performance verification.

Braille transcription is the process of converting information written in ordinary characters, such as books, documents and webpages, into a braille system. But Korean braille rules include various exceptions, including spacing, symbols and foreign-language notation, making accurate automatic transcription difficult.

Existing braille transcription programs used by visually impaired people convert characters and symbols based on simple rules. Errors have occurred in handling complex rules, such as mixed multilingual expressions including English and Korean, composite unit symbols and spacing around parentheses.

KAIST described K-Braille as a braille transcription system that understands sentences. While existing programs use substitution that simply replaces characters and symbols, K-Braille uses morphological analysis and sentence structure analysis (AST) to analyse structure and context, understand meaning and then convert it into braille. This allows it to handle more accurately various exception cases in revised braille rules, including sentences mixing foreign languages and Korean, complex symbol combinations and unit notation.

To verify accuracy, the team used the National Institute of Korean Language’s NLPAK, the country’s largest braille dataset, a print-braille parallel corpus. The data organises sentences pairing ordinary text with braille. The team extracted 17,943 sentences and conducted a full evaluation of how closely K-Braille’s output matched actual braille.

The practical compliance rate with braille rules was 100.0 percent. Morphological structure similarity, which shows how close braille sentence structure is to the correct answer, averaged 99.81 percent. In comparative verification using the same set of sentences against the National Institute of Korean Language’s official braille transcription program, Jumsarang 6.3.5.8, K-Braille showed a higher matching rate.

Kah, a researcher with a congenital severe visual impairment who led the study, said, "Braille is not simply a symbol for the visually impaired, but a language for reading the world." He said, "I hope this technology will further improve information accessibility for the visually impaired and serve as an opportunity to present a new technical benchmark for Korea’s braille translation AI field."

Keyword

#KAIST #K-Braille #National Institute of Korean Language #NLPAK #Jumsarang 6.3.5.8
Copyright © DigitalToday. All rights reserved. Unauthorized reproduction and redistribution are prohibited.