LLM models predict vocabulary difficulty with high accuracy

Researchers developed two models to predict vocabulary difficulty, with one achieving top results in a shared task. The high-accuracy model utilized a fine-tuned LLM with a soft-target loss function, reaching a correlation of over 0.91. An explainable model also demonstrated strong performance, correlating over 0.77, and provided insights into factors influencing word difficulty beyond just production ease, such as spelling complexity and test item construction. AI

IMPACT Demonstrates advanced LLM application in linguistic analysis and educational tooling.

RANK_REASON Academic paper detailing novel model development and evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• Adam Nohejl
• BEA 2026 Shared Task 1
• British Council
• Knowledge-based Vocabulary Lists (KVL)
• LLM
• Sakura

• paper
• product

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →