Federico Ranaldi

About

I am a third- and final-year PhD candidate in Data Science at the University of Rome Tor Vergata, supervised by Fabio Massimo Zanzotto, and a member of Human-Centric ART.

I'm currently a Visiting PhD student at the School of Informatics, University of Edinburgh, under the supervision of Shay B. Cohen, and a visiting member of the Cohort.

Research

I work at the intersection of NLP and Neuro-Symbolic AI. I study how Large Language Models acquire, represent and use knowledge, and how they can move beyond memorisation towards generalisation and abstract, structured reasoning. My research interests follow these directions:

• Grounding formal structures

  Code, logic, knowledge graphs and databases (relational or not): analyzing if LLMs can manipulate symbols abstractly, without relying on world knowledge that sometimes can lead them to errors (Syllogsisms are a typical testbed).

• Reasoning across languages, modalities and formal domains

  Multilingual, multimodal and agentic reasoning through new training and reinforcement-learning methods. While designing these methods, I like to draw inspiration from philosophy and cognitive science (e.g. Theory of Mind, Debate, reflection).

• Structured code generation across paradigms

  Transferring code-generation abilities across programming languages, whether or not they share paradigmatic constructs.

• Data contamination

  Detection and mitigation, especially for tasks that require complex reasoning rather than surface-level pattern matching.

I believe the future of AI lies in bridging the learning capabilities of LLMs with the principles of Symbolic AI, building systems that are reliable, cost-effective, compositional and interpretable by design.

Selected Publications

• 2026 EMNLP · Main Conference

  Surfacing Program Structure with Auxiliary Signals for Low-Resource Code Translation

• 2026 ACL · Main Conference

  Thinking in Schemas: Robust Syllogistic Reasoning in LLMs

• 2025 EMNLP · Main Conference

  Improving Multilingual Retrieval-Augmented Language Models through Dialectic Reasoning Argumentations

• 2025 Clic-It 2025 Italian Conference on Computational Linguistics

  Termite Italian Text-To-SQL (First Italian Text-To-SQL Benchmark ! xdxd )

• 2024 ACL · Findings

  Investigating the Impact of Data Contamination of Large Language Models in Text-to-SQL Translation

• 2023 EMNLP · Findings

  Exploring Linguistic Properties of Monolingual BERTs with Typological Classification among Languages

Projects