Yaraku Inc. (Headquarters: Shibuya, Tokyo; CEO: Suguru Sakanishi) , which operates the AI translation tool “YarakuZen” equipped with generative AI technologies such as ChatGPT, presented a case study of using YarakuZen in creative content at the international AI translation technology conference LANGUAGE INTELLIGENCE 2024, held in Vienna, Austria, on November 19 and 20.

 

Using LLM (Large Language Model) to Translate Artist Blogs into Seven Languages

At LANGUAGE INTELLIGENCE 2024, experts and representatives from companies around the world gathered to actively exchange information about the latest developments in AI-powered translation, interpretation, and speech processing technologies. From Yaraku, Hiroki Kudo, the Head of Overseas Strategy, took the stage to present a translation project in the entertainment industry utilizing YarakuZen. In this project, the blogs and video content of an artist on a world tour were translated into seven languages in under three hours using an LLM and post-editing techniques. The translation aimed to preserve the artist’s style and the nuances of the blog’s context while also considering cultural differences and the expectations of fans in different countries. One key factor was strictly following rules for translating proper nouns to maintain consistency and accuracy.

In this project, creative content that was difficult for traditional AI translation engines to handle was successfully translated by training ChatGPT-4o on previously translated phrases by human translators, and then refining the translations with further post-editing by translators. This approach led to highly accurate translation results. For example, the abbreviation form of “World Tour” in Japanese (ワルツ) was being incorrectly translated to “Waltz” referring to a music genre or dance style. However, after post-editing it was quickly translated correctly as “World Tour” moving forward. The post-editing process significantly reduced the revision rate from the typical 30-70% seen with traditional translation engines to only about 10%.

 

Rising Demand for Translation in the Entertainment Industry and Future Challenges—Volume and Speed are Key

This project demonstrated how LLM-based multilingual translation of blogs enabled the timely release of content to global fans, ensuring that the excitement from live events was maintained and fostering empathy, which ultimately led to stronger fan engagement on a global scale.

To support an artist’s international expansion, it is crucial to express their unique identity accurately in multiple languages. As such, the entertainment industry has a high demand for translations powered by LLMs. However, while this project successfully handled around 1,500 characters of text, there remain challenges in quickly translating large volumes of text. Yaraku believes that advancements in technology will eventually overcome these challenges.

 

LANGUAGE INTELLIGENCE 2024 https://www.lt-innovate.org/