In recent years, as technology becomes increasingly advanced and complex and customer needs continue to diversify, YAMAGATA has renewed its focus on the importance of creating manuals that are truly easy to understand and useful for our clients. As part of our efforts to produce even higher-quality manuals, we launched a joint research project on "Next-Generation Manuals" in 2023 with the Harmonious Engineering Laboratory of the Graduate School of Information Science and Technology at Hokkaido University, which possesses extensive expertise in natural language processing and AI technology.

On August 1, 2024, we held a "Paper Presentation and Lecture Session" at the YAMAGATA head office concerning our joint research project with Hokkaido University. This session was held for our employees with the goals of reporting on the results of the "Next-Generation Manuals" joint research, which has been ongoing since last year, and deepening their knowledge of AI technology. We were honored to welcome Associate Professor Tomohisa Yamashita, Assistant Professor Soichiro Yokoyama, and Mr. Uemae, a graduate student participating in this research, from Hokkaido University as guest lecturers.

Lecturers

Harmonious Engineering Laboratory, Graduate School of Information Science and Technology, Hokkaido University
Official Website of the Harmonious Engineering Laboratory

Associate Professor Tomohisa Yamashita
Assistant Professor Soichiro Yokoyama
Mr. Ryosuke Uemae

Report on Last Fiscal Year's Research Activities

First, our project members explained the content of last year's joint research activities. In the last fiscal year, with the cooperation of our in-house writers, we collected a dataset of poorly written sentences and their corrected versions. Project members and AI then classified and tagged this data, identifying which writing rules were applied to revise the poor text into the corrected text. Based on this information, Mr. Uemae of the Harmonious Engineering Laboratory took the lead in designing a system, creating a module for AI-powered technical writing, and conducting experiments. The quality of the AI-generated text was evaluated through a survey of just under 120 YAMAGATA employees.

Paper Presentation: Text Generation Using Large Language Models

First, Associate Professor Yamashita outlined the plan for automating the entire manual creation process and provided an overview of the plan for the second year of this joint research. The overall manual creation process was broken down into seven tasks, such as "Objective Setting" and "Text Authoring," and methods for handling each task with AI were considered. Currently, a path toward AI automation is visible for tasks from "Information Organization" to "Document Structuring," but tasks like "Objective Setting" and "Material Gathering" remain challenging. The plan is to proceed with AI automation for the feasible parts, with the ultimate goal of having AI complete the entire manual creation process. In the current fiscal year, the second year of the research, the specific goal is the "systemization of the text generation function," with plans to examine system challenges and approaches while keeping the automation of manual creation in mind.

Following that, Mr. Uemae gave a presentation on his graduate thesis, which was written based on the content above. His paper, on the theme of "Text Generation Using Large Language Models," explained the following three points in detail.

1. Objective of the Research:
   To create manuals that anyone can easily use by leveraging AI, with a particular focus on researching methods to revise human-written text into highly readable text.

2. Proposed Method:
   A clear explanation, using diagrams, of a writing rule application method consisting of a decision module and a correction module that use a large language model.

3. Experiments and Results:
   A report on the content and results of experiments that actually used ChatGPT to revise sentences.

Introduction to Research at the Harmonious Engineering Laboratory

Following the paper presentation, Associate Professor Yamashita and Assistant Professor Yokoyama introduced two AI-related research topics currently underway at the Harmonious Engineering Laboratory.

1. Document Extraction from Disaster Verification Reports:
   (Associate Professor Yamashita) The development of AI technology to efficiently extract necessary information from disaster verification reports created by local governments in the aftermath of a disaster.

2. Information Extraction from the Minutes of the Imperial Diet:
   (Assistant Professor Yokoyama) The development of a tool that uses AI to extract and analyze necessary information from the minutes of the Imperial Diet from the Meiji and Taisho eras.

Conclusion

This lecture session was a valuable opportunity for the participating employees to deepen their understanding of the progress of our joint research with Hokkaido University and the potential of AI technology. Many attendees commented that because the session introduced the application of AI to the field of manual creation support—an area many of them may be involved in professionally—they were able to view AI technology as a more immediate and relevant issue.

YAMAGATA has already established an environment where all employees can use multiple generative AIs, and we have held workshops on prompt engineering. In this joint research with Hokkaido University, we are leveraging the know-how gained through practical experiments to begin operating an AI-powered technical tool. This tool is not only realizing greater efficiency in technical writing tasks but is also being used in a wide range of operations, such as the pre-editing process for machine translation. This initiative, in which we rapidly elevated research outcomes into a practical, company-wide tool, has become a great source of confidence for YAMAGATA as a success story of open innovation through industry-academia collaboration.

The research results from fiscal year 2023 reported in this session are scheduled to be presented at the TC Symposium 2024, which will be held from Wednesday, October 9, to Friday, October 11, 2024, at the Kyoto Research Park.

YAMAGATA will continue to enhance employees' knowledge of AI technology and will provide ongoing updates on the progress of the "Next-Generation Manuals" project.