Japan is actively developing its own versions of ChatGPT, the popular artificial intelligence (AI) chatbot created by the US-based company OpenAI. This endeavor involves substantial investments from the Japanese government and leading technology firms like NEC, Fujitsu, and SoftBank, with the aim of building AI systems based on large language models (LLMs) specifically for the Japanese language.
Large language models like ChatGPT, such as GPT-3, are primarily trained on English language data, making them less effective for languages like Japanese due to differences in the alphabet system, limited data availability, and other factors, as stated by Keisuke Sakaguchi, a natural language processing specialist at Tohoku University.
Japanese, being a complex language with distinct structural differences from English, presents unique challenges for AI systems. ChatGPT often has to translate Japanese queries into English, find answers, and then translate responses back into Japanese. This can lead to errors, including the generation of rare or unknown characters.
For AI systems to be valuable and commercially viable in Japan, they need to accurately capture the language’s intricacies and cultural context. The structure of Japanese sentences differs significantly from English, making accurate and culturally appropriate responses crucial. Current LLMs may struggle to generate culturally sensitive and contextually accurate Japanese responses.
In response to these challenges, Japanese researchers and institutions are investing in creating their own Japanese-language LLMs. These efforts include utilizing Japan’s supercomputer, Fugaku, one of the world’s fastest, for training on Japanese-language input. This collaborative project involves the Tokyo Institute of Technology, Tohoku University, Fujitsu, and the government-funded RIKEN group of research centers. The resulting LLM is expected to be released next year and will be open source, unlike proprietary models like GPT-4.
Additionally, the Japanese Ministry of Education, Culture, Sports, Science and Technology is funding the development of a Japanese AI program tailored to scientific needs. This program aims to generate scientific hypotheses by learning from published research, potentially using at least 100 billion parameters.
Japanese companies are also actively commercializing their own LLM technologies. NEC has been using its generative AI for internal purposes, reducing the time required to create reports and software source code. SoftBank is investing in generative AI trained on Japanese text and plans to launch its own LLM in the near future.
Overall, Japan’s efforts to develop precise and effective AI chatbots in Japanese could lead to advancements in language processing and bridge the gap between Japan and the global AI research community. A successful Japanese version of ChatGPT is expected to have positive impacts on learning Japanese and facilitating international joint research.
