In a context where Europe is investing in more sovereign, reliable, and transparent artificial intelligence, initiatives such as DEEL and FOR, led by IRT Saint Exupéry, are playing a central role. Interpreto is an open-source library dedicated to NLP explainability, designed to better understand and interpret natural language processing models.
These structuring programs demonstrate their ability to effectively bring together research laboratories and companies to produce concrete tools with strong technological impact.
It is within this framework that Interpreto was designed and developed: an open-source Python library dedicated to the explainability of natural language processing (NLP) models, resulting from a collaborative effort led by IRT Saint Exupéry.
Interpreto perfectly illustrates the federating role of IRT Saint Exupéry:
- The creation of a fertile collaboration framework between academic and industrial stakeholders,
- Alignment with strategic challenges related to trustworthy AI,
- The delivery of an accessible, reusable, open-source tool.
The project brought together leading industrial and academic players around a shared objective, who co-developed a solution designed for real-world NLP explainability use cases.
Interpreto: an open-source library for NLP explainability
- Attribution-based and concept-based methods, addressing a wide range of interpretation needs,
- A set of metrics to evaluate the quality of explanations,
- Direct compatibility with Hugging Face models,
- Simple installation:
pip install interpreto
NLP explainability at the heart of a collaborative project led by IRT Saint Exupéry
The project was scientifically led by IRT Saint Exupéry and ANITI, with an interdisciplinary team bringing together industrial partners (Ampère, Renault, Thales, Scienta Lab) and academic partners (CentraleSupélec, IRIT, Khoury College of Computer Sciences, DataIA).
To learn more:
- Source code: https://github.com/FOR-sight-ai/interpreto
- Documentation: https://for-sight-ai.github.io/interpreto
- Scientific paper: https://arxiv.org/abs/2512.09730