Towards Conflict-Aware LLM Reasoning Grounded in Open Multilingual Knowledge Graphs

Large Language Models (LLMs) increasingly operate in settings where they must reason over heterogeneous sources of knowledgecontaining ambiguities and conflicts whose detection and interpretation requires implicit semantic understanding. Knowledge graphs provide a structured and formally defined representation of such knowledge, making them particularly valuable for studying and supporting these reasoning processes. My research investigates how well LLM-based systems detect and reason over conflicts grounded in formal knowledge graph semantics.

To this end, I am developing methodologies and benchmarks for systematically evaluating how models’ reasoning is affected by the availability of semantic context, variation in the underlying domain knowledge and semantic structures, and the language in which that knowledge is expressed. This includes studying situations in which the information needed to identify a contradiction is not explicitly provided and must instead be recovered or inferred by the model. The broader goal is to design more robust and trustworthy LLM-based systems that can reason over complex and conflicting knowledge. Because LLMs are increasingly used to support knowledge graph construction, validation, and completion tasks, improving their reasoning capabilities has the potential to enhance both the quality of knowledge graphs and the robustness of the applications built upon them.

Bio

Laura Balbi is a PhD student in Computer Science at the Faculty of Sciences of the University of Lisbon and a researcher at LASIGE. Her research lies at the intersection of Knowledge Graphs, Graph Neural Networks, and Neurosymbolic AI, with a particular focus on conflict-aware learning and trustworthy AI. Some of her current work investigates how effectively LLM-based systems can detect and reason over conflicts grounded in formal knowledge graph semantics, as part of a broader research agenda on developing AI systems that remain reliable in the presence of disagreement, uncertainty and context-dependent knowledge. She has contributed to European research projects exploring how structured knowledge can support explainable and knowledge-driven AI in biomedicine.