Innovation dans la compétence écrite pour l'ingénieur : analyse d'erreurs en FLE/FOS et pistes vers le diagnostic par l'IA

The primary objective of this presentation is to propose a typological analysis of written errors made by engineering students in a Spanish-speaking university context. By utilizing accessible digital tools for data processing, this study aims to identify linguistic obstacles and frequency patterns to improve pedagogical interventions, while establishing a structured dataset for future AI-driven diagnostic applications. Content details: Writing is often considered the most complex skill to master in French as a Foreign Language (FFL), particularly in engineering sciences where stylistic precision and logical structure are paramount. This difficulty arises from the simultaneous management of morphosyntax, technical vocabulary, and text cohesion. For Spanish-speaking learners, writing involves specific cognitive efforts to overcome L1-L2 interferences and the "fossilization" of errors, often exacerbated by a modern context of hyperconnectivity and assessment-driven learning. Our methodology is based on a representative corpus of 150 texts (50 per CEFR level: A1, A2, B1) collected at the Universidad Politécnica de Madrid. To process this data, we implemented a practical digital workflow: original handwritten texts were digitized using Google Lens (OCR), and the identified errors were systematically categorized in Excel spreadsheets. This spreadsheet-based analysis allows for the calculation of error frequencies and averages, providing a clear map of linguistic obstacles. The presentation will detail the academic context, the methodology, and the preliminary results of the study. Furthermore, we will discuss how this structured database serves as a foundational step toward the future integration of Artificial Intelligence (AI). We will conclude by outlining how such data can eventually train AI models to provide automated, personalized feedback for future-ready engineers.