英語口說精熟度之自動化評測技術研究

Abstract

none

The surge in English Medium Instruction (EMI) in higher education across Taiwan aims to prepare students for a competitive international environment. However, this shift introduces challenges, as students must grasp complex academic concepts in English, a non-native language, which may misrepresent their academic capabilities. Furthermore, instructors face difficulties discerning whether students' learning obstacles stem from language barriers or a lack of subject understanding. Addressing these concerns, we aim to develop a tailored Automated Speaking Assessment (ASA) system, with a focus to Taiwanese students. Our system emphasizes the unique linguistic and academic requirements of Taiwanese EMI settings. We investigate several models including traditional feature-based machine learning models and large pre-trained models, specifically fine-tuned with a Taiwanese EMI-focused dataset.Also, we propose innovative approaches to overcome the scarcity of relevant datasets with prototypical networks and address the issue of data imbalance via loss reweighting technique. By aligning assessment techniques closely with the specific needs of Taiwanese EMI students, our ASA system offers a more effective and contextually appropriate tool for language proficiency assessment in academic settings. The results of the experiments show the effectiveness of the methodologies.