Research
Our research centers on Human–AI Collaboration: helping people make better decisions, create, and learn with AI.
Knowledge-centric AI & NLP
How can we build AI-ready knowledge while preserving the meaning and evidence in domain documents?
We extract concepts, relationships, and conditions from documents and organize them into knowledge resources linked to their sources. By designing knowledge representations alongside retrieval and reasoning methods, we build foundations for AI that supports human judgment and work in specialized domains.
Current focus
Our current work structures laws and regulations governing broadcast content production and evaluates regulatory retrieval. We plan to extend these methods to knowledge structuring and question answering over financial documents.
Pluralistic Alignment
How can AI understand cultural context and reflect diverse perspectives throughout its reasoning?
We connect humanistic understanding of culture and social context with AI training and evaluation. We examine how models understand and reason about diverse cultures and values, with the aim of supporting collaboration with people from different backgrounds.
Current focus
We analyze and evaluate how large reasoning models (LRMs) interpret cultural contexts and form judgments.
By examining model-generated reasoning, we identify where cultural context is omitted or distorted and develop training and evaluation methods to address these issues. Process reward modeling is one representative approach.
Human-AI Co-Creativity
How can AI’s narrative planning capabilities help people turn their ideas into compelling works?
We develop generative models, agents, and interaction methods that support idea exploration, plot design, expression, and revision. Focusing initially on storytelling and writing, we aim to advance AI’s creative capabilities while enabling people to shape the creative process.
Current focus
Our first goal is to design convincing plot twists while maintaining foreshadowing and causal coherence. Building on this capability, we study collaborative writing environments in which novice writers and the public can explore and revise plots that reflect their intentions.
AI for Language Learning & Assessment
How can AI assessment and feedback improve learners’ actual language abilities?
We study the validity and consistency of assessment, the usefulness of feedback, and learning outcomes in L2 English speaking and writing. Our goal is adaptive scaffolding that responds to learners’ proficiency and progress as they practice and revise.
Current focus
We develop personalized feedback and methods for evaluating and validating AI-based assessment. Beyond improving individual pieces of work, we focus on helping learners develop the ability to perform independently.