PhD Position – Scientific Claim Verification in Research Articles

Type de contrat : CDD Contexte et atouts du poste Within the framework of a partnership (you can choose between) • public with French National Research Agency (ANR) Is regular travel foreseen for this post ? "travel expenses are covered within the limits of the scale in force". Mission confiée Context Large language models (LLMs) are increasingly used to assist scientific writing. While they can improve clarity and fluency, they may also introduce unsupported claims, distort results, or cite evidence in misleading ways. Detecting such problems requires more than analyzing the text alone: scientific claims must be checked against the evidence that supports them, which may appear in text, tables, figures, or cited papers. Scientific claim verification has received growing attention in recent years. Early benchmarks such as SciFact introduced the task of retrieving evidence from scientific papers and determining whether it supports or refutes a claim, while later work extended this task to the more realistic and challenging multimodal setting of scientific papers, where evidence may appear in tables and figures as well as in text. Yet, major challenges remain. Relevant evidence may be scattered across text, tables, figures, equations, code, and citations, and a correct prediction does not guarantee that a model relied on the right evidence. Developing methods that can both identify the appropriate evidence and verify claims reliably, while explaining their decisions and providing useful feedback to the user, is therefore an important step toward safer AI-assisted scientific writing. Objective The objective of this PhD is to develop and evaluate methods for verifying scientific claims against multimodal evidence in research articles. The work will address three main research questions, following a find → verify → explain pipeline: • Find - How can we identify the evidence relevant to a scientific claim across text, tables, figures, citations, and other modalities? • Verify - How can we determine whether a claim is supported by this evidence, while ensuring that the decision is grounded in the right evidence? • Explain - How can verification results be explained to authors and turned into useful feedback for revising scientific claims? Principales activités The PhD is expected to produce: • an open benchmark for multimodal scientific claim verification; • methods for identifying and aligning claims with supporting evidence; • evaluation protocols that assess not only prediction accuracy, but also robustness and evidence use; • a better understanding of the strengths and limitations of models for verifying scientific content. Compétences Technical skills and level required : • Master's degree (or 5-year university degree) in Computer Science, Computational Linguistics, AI/NLP, or a closely related field • Solid grounding in machine learning / deep learning; prior exposure to NLP is expected, multimodal learning (text + tables/figures) is a plus • Proficient in Python and standard deep learning frameworks (PyTorch, Hugging Face Transformers) • Ability to design experiments, build/annotate datasets, and critically analyze model behavior • Comfortable reading and producing scientific writing (papers, technical reports) Languages : • English: fluent, spoken and written (required for scientific publishing and daily research work) • French: not required to apply; willingness to learn is appreciated for integration into the lab and project Relational skills : • Able to work autonomously while contributing to a collaborative, interdisciplinary team (this PhD is embedded in the ANR Fabuleux project) • Clear communicator, able to explain technical results to researchers from different backgrounds • Good listener, open to feedback, comfortable with iterative/exploratory research Other valued appreciated : • Experience with claim verification, fact-checking, or scientific fact/evidence retrieval • Familiarity with retrieval-augmented generation (RAG) or multimodal document understanding • Prior research internship, publication, or open-source contribution in NLP/ML • Interest in AI safety / research integrity / responsible use of LLMs in science Inria, l'institut national de recherche dans les sciences et technologies du numérique, est en appui de l’État pour les stratégies nationales de recherche et d’innovation du numérique en tant qu'Agence de programmes. Inria mène plus de 300 projets de recherche et d’innovation avec ses 3500 scientifiques, ingénieurs et personnels d’appui, en partenariat avec les universités et l’écosystème numérique (entreprises, entrepreneurs, acteurs publics). Ensemble, nous explorons des domaines clés comme l'intelligence artificielle, la cybersécurité, l’informatique quantique, le Cloud, la transformation numérique de la santé, les jumeaux numériques ou encore les technologies numériques pour la défense. Nous construisons des solutions concrètes telles que des logiciels, des startups techn