NLP · Conversational AI · Dialogue Systems
Hi! I’m Dawn, a PhD candidate at Utrecht University in the Software Technologies for Learning and Teaching group, supervised by Johan Jeuring and Albert Gatt. I hold an MSc in Computer Science (Artificial Intelligence) from Delft University of Technology.
My research focuses on conversational AI and dialogue systems for professional training simulations, particularly how LLMs can generate, evaluate, and provide feedback during role-playing interactions. My PhD project is conducted in collaboration with DialogueTrainer, with whom I am also affiliated. My work investigates how these systems can be designed to be reliable, effective, and intuitive for real-world settings.
The 27th Meeting of the Special Interest Group on Discourse and Dialogue (SIGDIAL 2026) [Accepted (camera-ready in preparation)]
Description: We investigate whether role-playing LLM agents remain aligned with their intended conversational goals when users introduce competing distractions. Through controlled multi-turn simulations, we show a trade-off between conversational responsiveness and goal adherence.
The 18th International Natural Language Generation Conference (INLG 2025) [Slides] [Poster]
Description: We evaluate LLM-generated responses compared to human-authored responses in role-playing multi-turn interactions. We observed multi-turn degradation and consistency issues in LLM-generated responses.
Description: We extend a counselling-training agent with an Emotion-BDI model that can simulate dynamic user emotional states to improve realism and engagement in simulated dialogues.
Vol. 66, Computer Speech & Language
Description: We introduce a loss that uses label structure when training sentence encoders for text classification, so representations better reflect supervisory signal from the label space.
Nyan Cat © PRguitarman
