The Asahi Shimbun Company, one of Japan’s oldest and largest news organisations, is using technology to change its journalistic production. 

The company employs 3,800 people, operates 27 overseas bureaus, and circulates more than 4 million newspapers daily in its morning and evening editions. Online, it records about 100 million monthly pageviews and 30 million unique users, offering breaking news, in-depth reporting, podcasts, and specialised vertical media.

The transformation is being led by news company’s Media R&D, which was established in 2021. There are 20 core researchers and additional contributors across the country whose goal is to keep research closely tied to newsroom practice and social needs so that technological advances create tangible value.

“Our goal is to deliver value through advanced technologies,” said Hiroo Kusaba, deputy manager at Media R&D Center, at the recent INMA Asia/Pacific News Media Summit. “Our activities fall into three categories: product development, data journalism projects, and fundamental research in AI, including natural language, speech, and image processing.”

One of the projects developed by this team is “Invisible Intersections,” which illustrates how data-driven reporting can affect public policy. By analysing 1.6 million traffic accident records released by the National Police Agency, the team identified risks that had long gone unnoticed.

“And many accidents happened at small intersections, places without traffic lights and without official names,” he said. “Because these locations were not included in official statistics. Effective safety measures have not been implemented in some areas.”

The analysis also revealed striking differences between the timing of ordinary and fatal accidents. While the majority of crashes happened during daytime hours, deadly collisions peaked in the late afternoon, especially during winter months when twilight reduced visibility.

“From these insights, we produced feature articles and serialised stories, and the work had real impact,” he said. “Local governments responsible for road safety took action by installing new traffic signs and improving accident reporting methods. In this way, I believe the project helped address social issues. This contribution was recognised in 2023.”

Among the R&D Center’s most prominent innovations is Typoless, a Web-based proofreading application for Japanese. The tool reaches accuracy rates of 80% to 85% in internal testing, catching issues such as conversion errors and particle misuse. What sets it apart is its integration of Asahi Shimbun’s editorial standards, ensuring corrections meet professional newsroom expectations.

“It incorporates our company’s editorial rules to ensure professional-level accuracy,” he said. “In addition, it uses a large language model to check for potential content risks, such as wording that could trigger social media backlash.”

The project evolved from years of research in natural language processing, beginning in 2013. A patent for the core technology was secured in 2019, and a public prototype released in 2022 attracted attention, including promotion by a popular influencer on social media.

Typoless was officially launched externally in October 2023 and is now used by more than 100 companies and individual clients. The company plans to integrate the tool into its own newsroom production systems later this year.

According to Kusaba, “that wave of user engagement gave us the confidence to refine the product. By October 2023, Typoless was launched externally, and today it is used by more than 100 companies and individual clients. We are also preparing to integrate it into our own newsroom production systems.”

A second innovation, ALOFA, is a transcription tool designed to reduce one of journalism’s most time-consuming tasks. Built on the company’s proprietary speech recognition model, it cuts transcription time by about 60% while maintaining high accuracy.

“ALOFA was designed closely with journalists to fit newsroom workflows and includes features like automatic summaries and tables of contents to help reporters quickly navigate long recordings,” Kusaba explained.

The project originated in 2018 as part of a newsroom digital transformation initiative. A robust data pipeline was built to improve accuracy, and early testing was concentrated within the company.

Over the years, the system evolved, undergoing a major interface redesign in 2022 and expanding in scope. By 2024, it had more than 500 users internally and was processing over 2,000 hours of audio each month. In March this year, ALOFA was released as a public beta, with ongoing refinements driven by both internal and external feedback.

“Although Typeless and ALOFA started from different needs, Typoless from research in natural language processing and ALOFA from immediate newsroom needs, they both improved step by step through user feedback,” he said. “Our R&D works closely with the field to ensure technology meets real-world demands and delivers value to both the newsroom and society.”

The Media R&D Center keeps technology closely connected to practical use. It does not conduct research in isolation. The center focuses on projects that meet newsroom demands and address broader social issues.

This approach has allowed Asahi Shimbun to strengthen its journalism and extend its influence beyond the news industry, providing tools used by companies, institutions, and local governments, Kusaba said: “Technology alone is insufficient. What matters most is how it is delivered and how end users are involved.”

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .