Matteo Bortoletto

Research Interests

My research focuses on how AI can understand others' beliefs, goals, preferences, and intentions — a concept known as theory of mind — and how this understanding can improve interactions between humans and AI. In particular, I work on:

• Designing and assessing AI models for theory of mind tasks that span multiple modalities, including vision and natural language [AAAI'24] [ACL'24] [ECAI'24] [IJCAI-W'25]
• Probing and manipulating models' internal representations of mental states [EMNLP-F'25]
• Stress-testing theory of mind on large foundation models [EMNLP'25]
• Using theory of mind for action selection and intervetion [ArXiv]

Publications

Preprints

ProToM: Promoting Prosocial Behaviour via Theory of Mind-Informed Feedback
M Bortoletto, Y Zhou, L Ying, T Shu, A Bulling
arXiv:2509.05091 [ArXiv] [Website] [Code]

Unsupervised Partner Design Enables Robust Ad-hoc Teamwork
C Ruhdorfer, M Bortoletto, V Oei, A Penzkofer, A Bulling
arXiv:2508.06336 [ArXiv] [Code]

Peer Reviewed

ToM-SSI: Evaluating Theory of Mind in Situated Social Interactions
M Bortoletto, C Ruhdorfer, A Bulling
EMNLP 2025 [ArXiv] [Code] [Poster]

Brittle Minds, Fixable Activations: Understanding Belief Representations in Language Models
M Bortoletto, C Ruhdorfer, L Shi, A Bulling
EMNLP 2025 Findings [ArXiv] [Code] [Poster]
ICML 2024 Workshop on Mechanistic Interpretability [OpenReview]

The Overcooked Generalisation Challenge
C Ruhdorfer, M Bortoletto, A Penzkofer, A Bulling
TMLR [ArXiv] [Code]

The Yokai Learning Environment: Tracking Beliefs Over Space and Time
C Ruhdorfer, M Bortoletto, A Bulling
IJCAI 2025 Workshop on Theory of Mind (Oral) [ArXiv] [Code]

Explicit Modelling of Theory of Mind for Belief Prediction in Nonverbal Social Interactions
M Bortoletto, C Ruhdorfer, L Shi, A Bulling
ECAI 2024 (Oral) [ArXiv] [Code]

Limits of Theory of Mind Modelling in Dialogue-Based Collaborative Plan Acquisition
M Bortoletto, C Ruhdorfer, A Abdessaied, L Shi, A Bulling
ACL 2024 [Proceedings] [Code] [Poster]

Neural Reasoning about Agents' Goals, Preferences, and Actions
M Bortoletto, L Shi, A Bulling
AAAI 2024 [ArXiv] [Code] [Poster]

Improving Neural Saliency Prediction with a Cognitive Model of Human Visual Attention
E Sood, L Shi, M Bortoletto, Y Wang, P Müller, A Bulling
CogSci 2023 [Proceedings]

Exploring Natural Language Processing Methods for Interactive Behaviour Modelling
G Zhang, M Bortoletto, Z Hu, L Shi, M Bâce, A Bulling
INTERACT 2023 [ArXiv] [Code]

Contact

University: matteo [dot] bortoletto [at] uni-stuttgart [dot] de

Personal: mabo22 [dot] lab [at] gmail [dot] com