Lost in translation no more: How AI is bridging the language gap for international students in Korea
이 글자크기로 변경됩니다.
(예시) 가장 빠른 뉴스가 있고 다양한 정보, 쌍방향 소통이 숨쉬는 다음뉴스를 만나보세요. 다음뉴스는 국내외 주요이슈와 실시간 속보, 문화생활 및 다양한 분야의 뉴스를 입체적으로 전달하고 있습니다.
Proficiency in Korean is considered essential for international students in university classrooms, though real-time AI translation tools can assist those who are not yet fluent.
On Oct. 24, around 30 Korean and international students attended a seminar at the Gwangju Institute of Science and Technology (GIST)'s School of Environment and Engineering, where they listened to a Korean lecture about rising sea levels.
Unlike in usual classrooms, the Korean lecture was translated into English in real-time, with the translated text displayed next to the screen.
GIST teaches core courses in each major in English, but seminars with external speakers are some of the few lectures offered in Korean.
Since students are required to attend a certain number of these weekly seminars to graduate, AunionAI — a startup founded by GIST professor Kim Hong-kook in October last year — developed a real-time translation system for classroom use.
“Around 10 percent of our university’s student body is made up of international students, but it’s not easy to invite external speakers to give lectures in English,” said Professor Kim. “We thought we could create a tool that allows lecturers to speak in Korean while translating the lectures into English for international students.”
The AunionAI translation system collects audio from the lecture, translating it using AI and refining it with context memory. This adjusted translation is then displayed on the lecture screen.
The translation system was first trialed at environment and engineering school seminars during the spring semester and has since expanded to seminars at the School of Mechanical and Robotics Engineering this fall semester.
“It’s really helpful because the translation is shown just slightly behind the lecturer’s speed, giving me time to absorb it,” said Soeung Sreynich, a student who attended the seminar.
Most lectures are translated from Korean to English, but translations can also be provided in Chinese and Vietnamese, with translation from other foreign languages to Korean also possible.
AunionAI’s model has a bilingual evaluation understudy (BLEU) score — a metric for assessing machine translation quality — of 35 to 40 percent, which is comparable to Facebook’s translation models. Professor Kim said that a BLEU score of 60 percent typically exceeds human translation quality.
AunionAI has been working to further improve translation accuracy. Korean terms for specific regions pose a challenge, as shown when a past seminar mentioned Buan-gun, using gun to refer to the Korean word for county. The Korean word for army also uses the same pronunciation of gun, with the lecture being translated as Buan army.
Although the translations are not perfect, AunionAI continuously monitors and fine-tunes its translation model using lecture scripts from the TED Talk database and data provided through the government’s AI Hub project, of which AunionAI is a part.
Another way to enhance translation accuracy is by training the model with lecture materials provided in advance.
“If lecturers give us their materials before class, we can use that text to fine-tune the model, ensuring it shows terms and words relevant to the lectures,” Professor Kim said.
One advantage of AunionAI’s real-time model is that it begins translating within one or two seconds after the speaker starts talking, unlike other translation tools that wait until the speaker has finished their sentence. The displayed translation often starts with a word or two and updates as the lecturer continues speaking.
As the number of international students grows, other universities are expressing interest in AunionAI’s real-time lecture translation. The startup presented a demonstration at the Gwangju-Jeonnam University Presidents’ Association meeting in June, where many university officials expressed interest in adopting it. AunionAI has also begun discussions with a Seoul university, though specifics are still being worked out.
The startup will also participate in CES 2025 alongside the Gwangju Information & Content Agency, exhibiting at the Korea Trade-Investment Promotion Agency’s joint booth.
While the convenience of English translations could theoretically slow down Korean language learning, Professor Kim views it as a support system that aids international students.
“Even with translation, students are still listening to the Korean lecture with English subtitles,” Professor Kim said. “This can actually help with learning Korean, as they can verify what they’ve heard against the translated text.”
BY LEE TAE-HEE [lee.taehee2@joongang.co.kr]
Copyright © 코리아중앙데일리. 무단전재 및 재배포 금지.
- Korean celebrities congratulate Lee 'Faker' Sang-hyeok, T1 on League of Legends world title
- Two-thirds of Koreans oppose sending weapons to Ukraine, poll suggests
- Yulhee files lawsuit against ex-husband FTIsland's Choi Min-hwan
- G-Dragon arrives at Incheon International Airport in Tesla Cybertruck
- Coffee with a view of… North Korea? Starbucks eyes border observatory opening.
- SBS penalized, producer removed after NewJeans flashes iPhone in performance
- 'It's a slaughter': Video claiming to show North Korean soldier captured in Kursk emerges
- Army officer awaiting promotion admits to killing woman, dismembering corpse
- Korean YouTuber arrested for latest assault on U.S. streamer Johnny Somali
- 'Culinary Class Wars' Triple Star under fire after allegations from 'exes'