Using PTE writing tasks and large language models (LLMs) for dynamic assessment

2025-2026

Project title: Bridging language assessment and learning: Using PTE writing tasks and large language models (LLMs) for dynamic assessment

Dynamic assessment (DA) integrates mediation into assessment procedures to reveal learners' emerging abilities within their Zone of Proximal Development. While DA has shown promise in second language (L2) writing contexts, its implementation has been limited primarily to small-scale settings. Interventionist approaches offer scalability through predetermined prompts but lack flexible, dialogic qualities, whereas interactionist approaches provide responsive mediation but remain resource-intensive. Meanwhile, standardized writing tasks from tests like PTE Academic have been underexplored as DA tools, and the integration of Large Language Models (LLMs) into DA remains largely unexamined despite the rapid advancement of generative AI technologies. This study addresses these gaps by repurposing PTE Academic writing tasks for DA purposes in test preparation contexts and exploring how LLMs can be fine-tuned to provide effective mediation.

Grounded in Vygotsky's sociocultural theory and Black and Wiliam's formative assessment framework, the research investigates three questions: (1) how LLMs can be fine-tuned for effective DA mediation, (2) whether PTE writing tasks with LLM mediation lead to improvements in L2 writing ability over time, and (3) learners' perceptions of this approach. The study will fine-tune an LLM and then implement a longitudinal design with 15 prospective PTE test takers participating in DA sessions over two months. Mixed-methods analysis will examine both L2 writing development and participant perceptions and experiences. This research contributes to learning-oriented assessment by demonstrating how standardized tasks can support formative learning, while establishing evidence-based guidelines for integrating LLMs into DA contexts.