Presented at ICAICTA2026
Applied Information Systems Laboratory
Observing data and models
The Applied Information Systems Laboratory studies natural language processing: how computers can understand and process the language we use every day (natural language). In particular, we are committed to observing data and models closely.
As an example, consider the task of answering a question written in natural language, such as “How many meters high is Mount Fuji?” (question answering). Today, the common approach is end-to-end: prepare a dataset of many questions and answers, and obtain a model that performs the task through machine learning (deep learning). However, this approach is known to often produce models that behave in ways people do not expect. In other words, rather than simply aiming for the final accuracy, it is important to ask why the task can be solved with these data and this model. For this reason, our research in natural language processing centers on methods for closely observing data and models.
News
- October 20, 2026Presentation
- August 18, 2026AwardHitachi, Ltd. Award, 21st Symposium of the Young Researcher Association for NLP Studies (YANS2026)
- August 17, 2026PresentationPresented 2 papers at the 21st Symposium of the Young Researcher Association for NLP Studies (YANS2026)
- May 15, 2026EventHeld a welcome party for new members
- March 10, 2026PresentationPresented at NLP2026
Research Topics
- Communication during disasters
How place-name notation in disaster-time social media posts relates to the quality of the information.
- Multilingual and multicultural language models, and social media analysis
Emotion in social media and news headlines, and cultural bias in LLMs, across Thai, Indonesian, Finnish and more.
- Evidence-based judgment: evaluating LLM reliability
Whether LLMs rely on the evidence they are given rather than on their internal knowledge.
- Do LLMs really understand?
Explaining the causes of accuracy drops and errors from inside the model.
- Finding style, emotion and stance inside language models
Discovering directions in model internals that correspond to such attributes, using sparse autoencoders.
Staff
- Masatoshi TSUCHIYA, Professor (Principal Investigator / Head of Laboratory)
- Zhuo Binggang, Assistant Professor
Address
Toyohashi University of Technology
1-1 Hibarigaoka, Tenpakucho, Toyohashi-shi, Aichi, 441-8580 Japan