Headwaters Co., Ltd. Announces Research Results on "Narrative Translation Sensitivity Surrogate by Large Language Models (LLMs)" at the 21st Annual Conference of the Japan Society of Kansei Engineering & ISASE2026
Headwaters Co., Ltd. presented its research on "Narrative Translation Sensitivity Surrogate by Large Language Models (LLMs)" at a major academic conference. This research introduces the concept of a "Kansei Surrogate," which uses AI like LLMs to proxy and complement subjective human emotional evaluations. Focusing on the aesthetic quality of narrative translation—including worldview, characterization, and tone—the study organized key considerations for designing LLM-based evaluation methods. This aims to address challenges in content localization, such as resource shortages and the difficulty of objectively assessing subjective qualities, paving the way for advanced AI-driven quality assessment in various creative fields.
📋 Article Processing Timeline
- 📰 Published: April 14, 2026 at 01:00
- 🔍 Collected: April 13, 2026 at 16:35
- 🤖 AI Analyzed: April 16, 2026 at 03:17 (58h 41m after Collected)
Headwaters Co., Ltd. (Headquarters: Shinjuku-ku, Tokyo; Representative Director: Yosuke Shinoda; hereinafter "Headwaters"), an AI solution business provider, announced that its members presented research results on "Narrative Translation Sensitivity Surrogate by Large Language Models (LLMs)" at the 21st Annual Conference of the Japan Society of Kansei Engineering & ISASE2026.
In this research, a "Kansei Surrogate" refers to a concept for substituting or complementing evaluations related to subjective and individually varying "human feelings (Kansei)" using AI such as LLMs.
Based on this concept, the research focused on the emotional quality of narrative translation, such as worldview, characterization, and tone, and organized the key points for designing evaluation using LLMs.
**The 21st Annual Conference of the Japan Society of Kansei Engineering & ISASE2026 / Conference & Presentation Overview**
* **Conference Overview:**
* Conference Name: The 21st Annual Conference of the Japan Society of Kansei Engineering & ISASE2026
* Dates: March 16 (Mon) - 18 (Wed), 2026
* Venue: Utsunomiya University Yoto Campus
* Official Page: https://www.jske.org/conference/jske21s/
* **Research Presentation Overview:**
* Title: Narrative Translation Sensitivity Surrogate by Large Language Models
* Presentation Date: March 16, 2026
* Presenters: Takayuki Shimotomai, Fuuka Shibata, Hiroki Shiga, Koki Takeishi (Headwaters Co., Ltd.), Miwako Kamijo (Sagami Women's University)
* Details: https://pub.confit.atlas.jp/ja/event/jske2026s/presentation/1D02-05
**Background: Expanding Content Translation and Challenges in Quality Evaluation**
With the global expansion of content such as games, anime, movies, and books, the demand for translation and localization is increasing significantly. On the other hand, challenges include a shortage of translation resources and the considerable human effort and cost involved in quality evaluation.
Especially in narrative translation, elements like the original worldview, character personality, tone, and dynamism of scenes greatly influence quality, in addition to semantic accuracy. However, such qualities are highly subjective and context-dependent, making them difficult to evaluate sufficiently with existing automated evaluation metrics.
Therefore, with the practical application of generative AI and LLMs progressing, how to incorporate emotional quality into the evaluation process has become a new technical challenge.
**Research Overview: Organizing Evaluation Design**
In the research, multiple axes important for narrative translation were set as evaluation perspectives: Fidelity, Worldview, Character, Tone, Dynamics, and Overall evaluation.
Furthermore, direct evaluation and comparative evaluation (pair comparison) were used for examination. By adopting the Bradley-Terry model for aggregating comparative evaluation results, a framework that easily structures subjective evaluation results was incorporated.
This research demonstrates the importance of defining evaluation axes first and then designing AI evaluation based on those axes when evaluating subjective quality.
**Future Developments**
The evaluation design organized in this research is expected to be applied not only to advanced AI translation quality evaluation but also to areas requiring subjective quality evaluation, such as generative AI-based text generation, localization, brand tone consistency checks, and content quality management.
Headwaters will continue to advance research and development of evaluation technologies and design knowledge that can be utilized in practical applications.
**About Trademarks**
Proper nouns such as events mentioned are trademarks or registered trademarks of their respective companies.
**Company Information**
Company Name: Headwaters Co., Ltd.
Location: 4F Shinjuku Island Tower, 6-5-1 Nishi-Shinjuku, Shinjuku-ku, Tokyo 163-1304
Representative: Yosuke Shinoda, Representative Director
Established: November 2005
URL: https://www.headwaters.co.jp/
In this research, a "Kansei Surrogate" refers to a concept for substituting or complementing evaluations related to subjective and individually varying "human feelings (Kansei)" using AI such as LLMs.
Based on this concept, the research focused on the emotional quality of narrative translation, such as worldview, characterization, and tone, and organized the key points for designing evaluation using LLMs.
**The 21st Annual Conference of the Japan Society of Kansei Engineering & ISASE2026 / Conference & Presentation Overview**
* **Conference Overview:**
* Conference Name: The 21st Annual Conference of the Japan Society of Kansei Engineering & ISASE2026
* Dates: March 16 (Mon) - 18 (Wed), 2026
* Venue: Utsunomiya University Yoto Campus
* Official Page: https://www.jske.org/conference/jske21s/
* **Research Presentation Overview:**
* Title: Narrative Translation Sensitivity Surrogate by Large Language Models
* Presentation Date: March 16, 2026
* Presenters: Takayuki Shimotomai, Fuuka Shibata, Hiroki Shiga, Koki Takeishi (Headwaters Co., Ltd.), Miwako Kamijo (Sagami Women's University)
* Details: https://pub.confit.atlas.jp/ja/event/jske2026s/presentation/1D02-05
**Background: Expanding Content Translation and Challenges in Quality Evaluation**
With the global expansion of content such as games, anime, movies, and books, the demand for translation and localization is increasing significantly. On the other hand, challenges include a shortage of translation resources and the considerable human effort and cost involved in quality evaluation.
Especially in narrative translation, elements like the original worldview, character personality, tone, and dynamism of scenes greatly influence quality, in addition to semantic accuracy. However, such qualities are highly subjective and context-dependent, making them difficult to evaluate sufficiently with existing automated evaluation metrics.
Therefore, with the practical application of generative AI and LLMs progressing, how to incorporate emotional quality into the evaluation process has become a new technical challenge.
**Research Overview: Organizing Evaluation Design**
In the research, multiple axes important for narrative translation were set as evaluation perspectives: Fidelity, Worldview, Character, Tone, Dynamics, and Overall evaluation.
Furthermore, direct evaluation and comparative evaluation (pair comparison) were used for examination. By adopting the Bradley-Terry model for aggregating comparative evaluation results, a framework that easily structures subjective evaluation results was incorporated.
This research demonstrates the importance of defining evaluation axes first and then designing AI evaluation based on those axes when evaluating subjective quality.
**Future Developments**
The evaluation design organized in this research is expected to be applied not only to advanced AI translation quality evaluation but also to areas requiring subjective quality evaluation, such as generative AI-based text generation, localization, brand tone consistency checks, and content quality management.
Headwaters will continue to advance research and development of evaluation technologies and design knowledge that can be utilized in practical applications.
**About Trademarks**
Proper nouns such as events mentioned are trademarks or registered trademarks of their respective companies.
**Company Information**
Company Name: Headwaters Co., Ltd.
Location: 4F Shinjuku Island Tower, 6-5-1 Nishi-Shinjuku, Shinjuku-ku, Tokyo 163-1304
Representative: Yosuke Shinoda, Representative Director
Established: November 2005
URL: https://www.headwaters.co.jp/