DEVELOPMENT OF AN ENGLISH ORAL ASSESSMENT SYSTEM WITH THE GEPT DATASET

In the rapidly evolving intersection of artificial intelligence and educational technology, multi-perspective assessments are transforming language learning. However, much of the existing research focuses on single-dimensional evaluations, often neglecting the potential of a comprehensive, multi-dimensional oral assessment system that addresses acoustic features, content, and language use simultaneously. Few studies have developed such systems based on established language proficiency tests like the General English Profi-ciency Test (GEPT). This study breaks new ground by creating an integrated, multi-dimensional oral assessment system tailored to the Intermediate level of the GEPT Speaking Test, specifically targeting the Picture Description section. We introduce two novel architectural innovations: the first integrates gram-mar error correction using GECToR and ERRANT into the Language Use Module, while the second enhances the Content Module by combining image and question analy-sis through Blip2 and large language models (LLMs). Our system demonstrates robust performance, achieving 71 % ac-curacy on familiar content and 68% accuracy on unfamiliar content.