Joyce Chai

Joyce Chai is a prominent academic researcher in computational linguistics and AI with extensive contributions to grounded language learning, embodied agents, and human-machine collaboration. Her work spans vision-language models, theory of mind implementation, and task guidance systems.

• Key research themes: Language grounding in physical/social contexts
• Embodied AI and situated reasoning
• Interactive learning frameworks
• Zero-shot and continual concept acquisition

Recent publications demonstrate her leadership in:

• Developing TRAVER for coding tutoring agents
• Creating W2W grounded language model
• Advancing theory of mind evaluation in LLMs
• Establishing HAR reasoning strategies for coherent physical reasoning

She has mentored numerous students including Ziqiao Ma, Shane Storks, and Yuwei Bao. Her work appears in top venues like ACL, EMNLP, and NAACL with focus on practical applications like cake-making guidance systems (WTaG) and autonomous driving dialogue (DOROTHIE).

Technical contributions include:

• MetaReVision retrieval-enhanced meta-learning
• EpiCA network for compositional concept recognition
• Neuro-symbolic DANLI agent architecture
• Pragmatic Rational Speaker framework