By Makoto Hattori

8:00 JST, December 20, 2025

AI translation has made significant strides in recent years, now handling tasks once considered uniquely human, such as translating manga and literary works, and providing simultaneous interpretation at international conferences. Schools have also begun introducing apps that correct English compositions and evaluate pronunciation. AI is rapidly becoming more significant to the learning of foreign languages and the ways they are taught.

Technology developed by manga-translation company Mantra symbolizes this shift. “To begin with, we upload the image data of an entire manga chapter to a computer — images in which Japanese text appears inside speech balloons,” explained Mantra’s Chief Technology Officer Ryota Hinami, 34. The system uses a process called optical character recognition to extract the text and then runs it through AI translation, after which professional human translators revise the output. Edited translation data are accumulated and reused to maintain consistent tone and terminology.

CEO Shonosuke Ishiwatari, 33, describes the core of their technology as follows: “A major feature of our system is that it automatically infers a list of character names from the images the AI reads, identifies who appears in each panel, and estimates which character each speech balloon belongs to.” This makes it possible to produce translations that reflect the personality of each character, driving a surge in demand. Over the past year, the number of pages processed per month has grown from 100,000 to 200,000, and Mantra’s work now extends to novels, anime and games. “In entertainment translation, preserving each character’s authenticity is crucial. The strengths we developed in manga translation apply directly to that,” Ishiwatari said.

Underlying this growth is the rapid global expansion of Japan’s content industries — games, anime and manga. With the spread of video streaming services, simultaneous worldwide releases have become commonplace, and hit titles are distributed in dozens of countries. Publishers, animation studios and game developers now plan for overseas markets from the outset, making translation speed and quality decisive business factors. This rising demand is fueling the development of entertainment-focused translation technologies like Mantra’s.

Google Translate launched in 2016, followed by DeepL in 2017. Both are forms of neural machine translation, which mimics the neural circuits of the human brain and learns from vast repositories of translation data. Soon after their debut, they reached a proficiency level comparable to a TOEIC English exam score in the........

© The Japan News