Skip to main content
The European High Performance Computing Joint Undertaking (EuroHPC JU)

A Self-Taught Framework for Explainable Second Language Assessment in Low-Resource Scenarios: A Case Study on Basque

78003 Awarded Resources (in node hours)
Leonardo BOOSTER System Partition
July 2026 - January 2027 Allocation Period

AI Technology: Natural Language Processing; Generative Language Modeling.

This project proposes a new methodology to develop Automatic Essay Scoring (AES) systems for the Basque low-resource language.

The methodology relies on a pair of interacting components: an 'expert model' that generates detailed essay evaluations, and a 'judge model' that filters these assessments to retain only pedagogically valuable feedback. 

The approach leverages the self-taught paradigm, requiring minimal pedagogical annotations, and aims to address the unequal treatment of languages in existing AES systems through transferable methodologies. 

The study's performance analysis will be grounded in a series of four experiments, each looking at different aspects of the proposed system. 

The intended outcomes include robust, pedagogically aligned AES models, as well as a sound methodology that can be transferred to other low-resource languages in Europe.

Principal Investigator, Institution and Country