When OpenAI released the ChatGPT chatbot in November 2022, artificial intelligence (AI) became one of the most talked about topics in the world. The implications for organizations are enormous and research according to Goldman Sachs indicates that AI has the potential to fuel a 7% increase in global gross domestic product (GDP), or almost $7 trillion, over the next decade.
But what does the world of AI translation hold for Japanese translations? In this article, we explore the development of AI software in Japanese, look at how AI is used in terms of AI for Japanese translation, and explore some of the regulatory and other challenges that come with such developments.
The development of AI software in the Japanese language: Accuracy and opportunities
There’s been much talk about AI over the past year with large language model (LLM) development becoming the talk of the town. Apart from ChatGPT, there’s Bard, Bing, and Deep L, among others that are making waves. But in terms of Japanese LLM, there’s still a lot of work that lies ahead. Some of the reasons behind this are issues of accuracy.
In a July 2023 article, the Japan Times conducted an experiment in which it placed four AI platforms next to each other, comparing them for accuracy in terms of translating from Japanese to English. The results showed that there were numerous challenges in translating some Japanese texts (including a speech, the national anthem, and an excerpt from a Nobel laureate’s work) with regard to accuracy. The main reason behind this is that LLM development, especially for Japanese, is still in its infancy. And there are several reasons for this, which we’ll cover in more detail below.
However, in terms of opportunities, AI for Japanese translation has a long road to cover but it has been said that with human input, can simplify and speed up translation processes, meaning greater efficiencies and more accurate output. Despite linguistic nuances that AI sometimes cannot cover, and the absence of strong LLM development for Japanese, there are ample opportunities for further developments in this area.
How are Japanese companies pursuing generative AI?
Although some companies in Japan have prohibited their employees from using generative AI, there are many companies and industries that have used AI prior to the launch of generative AI. These industries include finance, manufacturing, infrastructure, healthcare, and nursing care, as well as services such as inspection, maintenance, and call center operations.
These companies and organizations within industries are increasingly starting to apply generative AI in a variety of areas of work, including both routine work as well as work that has opened up new opportunities for innovation: from the creation of images of completed buildings to be constructed and coding and debugging.
Regulative measures in Japan regarding generative AI
According to sources, generative AI is being rapidly implemented in Japanese society. However, there are some risks that the Japanese government and various high-level policymakers are concerned about. Examples of these risks include, but are not limited to:
- Third-party copyright infringements
- Use of incorrect information
- Leaks or improper use of confidential and personal information
- Misuse of generative AI
Nevertheless, the Asian nation is currently focusing on implementing “soft” as opposed to “hard” laws such as the EU has done with generative AI, in order to ensure that innovation by AI is not impeded. This has been achieved by building a framework for governance and guidelines instead of imposing obligations through laws and regulations. As such, the private sector has been left to conduct the governance voluntarily. An example of such a guideline is the “Governance Guidelines for Implementation of AI Principles” published by the Ministry of Economy, Trade and Industry (METI) on 9 July 2021 (as amended in Version 1.1 of 28 January 2022).
However, calls have been made by Japan’s Liberal Democratic Party to develop a new national strategy and to review previous measures as soon as possible, with regard to the Japanese government’s existing AI strategy.
Challenges that generative AI poses and obstacles to full implementation in Japan
When it comes to the challenges posed by AI and the obstacles that Japan faces with regard to this new development, it’s important to consider what the country has to deal with in order to ensure greater transparency, fairness, and respect and upholding of human rights in relation to implementing generative AI on a wider scale.
Among the first challenges is the limited availability of a robust community of software engineers to develop the necessary infrastructure and applications. For example, Japan is expected to face a deficit of 789,000 software engineers by 2030.
Apart from software developers, there is also the challenge of hardware, as LLMs need to be trained using AI supercomputers. Currently, no private company in Japan possesses its own machine with the requisite capabilities. The alternative, of course, is to use government-controlled supercomputers in Japan’s pursuit of LLMs.
Furthermore, there is the challenge of ensuring accurate translations of the Japanese language itself. Japanese is a language in which sentences sometimes lack a subject and, as such, natural translations into English are often difficult to achieve. Literature, for instance, is one of the most difficult tasks for generative AI to translate, because it requires a translation of the nuances of a story in an eloquent manner and it requires a more in-depth understanding of the time period being described, which is something generative AI does not yet have the capacity of doing.
In addition to this, there are different prompts that can be given for the same task, producing different results.
Another issue is that generative AI tools can be weak in practical and detailed areas, such as explaining grammar and spelling and doing arithmetic.
Finally, AI chatbots are known for imagining facts and making reasoning errors. LLMs have been accused of being able to “hallucinate” and present inaccurate information as fact, which many have warned about.
Final thoughts
LLM development, and especially Japanese LLM require a lot more work for more accurate AI translation and more accurate and factually correct output. Although there are some legislative steps being taken towards a “soft” regulation of the industry, much remains unknown as to how generative AI will be regulated in Japan. What is clear is that AI for Japanese translation is essential if Japan wants to maintain its status as a high technologically developed country and as competition between the United States and the EU continues.