SWIM Evaluates LLM Capability for Student Writing Simulation
September 4, 2026
The SWIM framework tests whether LLMs can simulate student writing through proficiency-conditioned generation. Experiments show that while models can adjust content-oriented traits, they struggle to replicate specific lexical, grammatical, and organizational variations seen in different proficiency levels.
HOW THIS AFFECTS YOU
●
researcherCurrent prompting and fine-tuning methods are insufficient for precise control over the multidimensional linguistic traits of student writing.