Luc Pommeret
Research engineer, LISN (CNRS). Atomic propositions for verifiable AI.
Contact:
When an LLM generates a paragraph, how do you know each claim is true? How do neural networks represent semantics? I decompose natural language into atomic propositions, minimal, verifiable units of meaning, to build formally grounded verification systems. My work bridges Carnap and Bar-Hillel's semantic information theory with modern NLP.
Research
LISN (CNRS) — since April 2025, supervised by Sophie Rosset, Thomas Gerald, Christophe Servan and Sahar Ghannay. Atomic propositions in NLP: evaluation and improvement of RAG systems, training a propositioner. — internship report
IRIF — April–September 2024, supervised by Michel de Rougemont. Emergent capabilities in large language models. — internship report
Publications & talks
Pommeret, Gerald, Servan, Ghannay, Paroubek, Rosset. "LLM based Atomic Propositions help weak extractors: Evaluation of a Propositioner for triplet extraction". KG-LLM 2026, LREC 2026, Palma de Mallorca, May 16, 2026.
Pommeret, Wagret, Deret. "THIVLVC: Retrieval Augmented Dependency Parsing for Latin". LT4HALA 2026 / EvaLatin 2026, LREC 2026, Palma de Mallorca, May 11, 2026.
Pommeret, Gerald, Rosset, Servan, Ghannay. "Atomic Propositions: Bridging Neural and Symbolic Approaches for Interpretability by Design". GDR TAL Interpretability Day, ISIR Jussieu, March 17, 2026. Oral presentation.
Pommeret, Rosset, Servan, Ghannay. "AtomicEval: Evaluation Framework for Atomic Proposition Autonomy with French Propositioner". JDSE 2025, Paris-Saclay.
Pommeret. "Exploring Emergent Skills with Chess-GPT". PFIA 2024, La Rochelle.