As part of the Stockmark team, we won first place in the shared task at the Language Technologies for Crisis Preparedness and Response (LT4CPR) workshop at AACL-IJCNLP 2026. Team Stockmark ranked first in both the Primary System Ranking and the Best System Ranking.
I will speak at Encraft #26: Dissecting Ontologies (original Japanese title: Encraft #26 オントロジー解体新書), a Japanese-language event hosted by Knowledge Work on September 28, 2026. My talk is titled “Generating New Business Ideas with Ontologies” (original Japanese title: オントロジーで新規事業を発想する).
I will present at the 44th Web and Social Media Journal Club, an online event held on August 20, 2026 (original Japanese title: ウェブ・ソーシャルメディア論文読み会). I will introduce the ICWSM 2026 paper “Fluent but Unfeeling: The Emotional Blind Spots of Language Models.” (slides)
Our co-author Naoki Shindo presented “An Autonomous Exploration Agent for Generating Cross-Domain Application Ideas That Balance Novelty and Feasibility” (original Japanese title: 意外性と実現可能性を両立する異分野転用アイデア創出のための自律的探索エージェント; paper in Japanese) at the 21st Symposium for Young Researchers of the Association for Natural Language Processing (YANS 2026).
Our ACL 2026 paper, “Aggregate vs. Personalized Judges in Business Idea Evaluation: Evidence from Expert Disagreement,” was featured on the websites of Stockmark and AIST Solutions.
(In Japanese) I wrote an article on Stockmark’s tech blog titled “Are We Asking LLM-as-a-Judge for a Single Right Answer? (LLM-as-a-Judge に「唯一の正解」を求めすぎていないか？)”. It analyzes the diversity of expert evaluations in business idea assessment and discusses a human-centered evaluation design where humans decide which perspective to adopt and models handle the evaluation from that perspective. The post explains our paper accepted to the ACL 2026 Industry Track.
We published a paper proposing a model editing approach to align language models with expert judgment in subjective evaluation tasks. This is joint work with Lin, Ishigaki (AIST), Lee, and Chen. Why Expert Alignment Is Hard: Evidence from Subjective Evaluation. arXiv.
Our paper was accepted to the ACL 2026 Industry Track as an oral presentation. Aggregate vs. Personalized Judges in Business Idea Evaluation: Evidence from Expert Disagreement. arXiv.
I will serve as an organizer for the organized session OS-37 “Generative AI and Knowledge Graphs” at the 40th Annual Conference of JSAI (JSAI2026), to be held in June 2026. The session is co-organized with Prof. Kouji Kozaki (Osaka Electro-Communication University), Prof. Takeshi Morita (Aoyama Gakuin University), and Shigeri Kurokawa (KDDI Research). We look forward to your submissions and participation.
I am attending the 32nd Annual Meeting of the Association for Natural Language Processing (NLP2026), held in Nagasaki, March 2026. I also co-authored a blog post introducing the five research papers our team at Stockmark is presenting at the conference: Natural Language Processing Supporting Seeds-Needs Matching.