Maarten Sap

My research group explores the critical intersection of ethics, accountability, and human-centered AI in modern applications. We emphasize the importance of cultural sensitivity and bias mitigation by evaluating AI systems through a social lens. Key contributions include the paper [Mind the Gesture: Evaluating AI Sensitivity to Culturally Offensive Non-Verbal Gestures](https://arxiv.org/abs/2502.17710), which addresses the nuances of non-verbal communication in AI interactions. Additionally, our study on [Mitigating Bias in RAG: Controlling the Embedder](https://arxiv.org/abs/2502.17390) highlights strategies for minimizing biases inherent in AI model training. Furthermore, the insights from [Let Them Down Easy! Contextual Effects of LLM Guardrails on User Perceptions and Preferences](https://arxiv.org/abs/2506.00195) shed light on user perceptions and trust in AI systems with guardrails designed for safety.

My research group explores the rich complexity of narratives and the role they play in shaping human experiences and AI interactions. Our work investigates how stories influence empathy and can be leveraged in AI models. Key papers include [HEART-felt Narratives: Tracing Empathy and Narrative Style in Personal Stories with LLMs](https://arxiv.org/abs/2405.17633), which examines the stylistic features that evoke empathy in narratives. We also delve into the idea of narrative perceptions with the study [Quantifying the narrative flow of imagined versus autobiographical stories](https://www.pnas.org/doi/10.1073/pnas.2211715119), providing a framework to analyze differences in storytelling. Lastly, [The Empirical Variability of Narrative Perceptions of Social Media Texts](https://aclanthology.org/2024.emnlp-main.1113/) investigates how varying contexts change the interpretation of narratives across digital platforms.

My research group explores the development of AI agents that exhibit enhanced social intelligence and the ability to interact meaningfully with humans. A primary focus is on understanding how personality traits influence AI behaviors in conversations, as shown in the paper [BIG5-CHAT: Shaping LLM Personalities Through Training on Human-Grounded Data](http://arxiv.org/abs/2410.16491). We also investigate the effectiveness of AI agents in collaborative settings through [Interactive Agents to Overcome Ambiguity in Software Engineering](https://arxiv.org/abs/2502.13069), which discusses the dynamic adaptability of AI in complicated environments. Our exploration of negotiation dialogues in [Exploring Big Five Personality and AI Capability Effects in LLM-Simulated Negotiation Dialogues](https://arxiv.org/abs/2506.15928) highlights how AI can be tailored to user preferences for more effective interactions.

My research group explores methodologies for identifying and mitigating toxic language generated or perpetuated by AI systems. We focus on creating frameworks that not only detect toxicity but also promote healthier communication patterns. One significant contribution is the study [PolyGuard: A Multilingual Safety Moderation Tool for 17 Languages](https://arxiv.org/abs/2504.04377), which presents a robust mechanism for moderating harmful content across different languages. Another important paper, [Counterspeakers’ Perspectives: Unveiling Barriers and AI Needs in the Fight against Online Hate](https://arxiv.org/abs/2403.00179), identifies practical challenges faced by communities combating toxic language and biases. Furthermore, our research considers context in evaluations, exemplified in [COBRA Frames: Contextual Reasoning about Effects and Harms of Offensive Statements](http://arxiv.org/abs/2306.01985), which aims to develop a nuanced understanding of how toxic statements affect social dynamics.