Toward Beginner-Friendly LLMs for Language Learning: Controlling Difficulty in Conversation
Date:
Abstract
In this talk I discuss my work with Meiqing (Emma) Jin on controlling the output difficulty of LLMs to better suit beginner language learners in Japanese. We test both prompting, reranking, and modular difficulty control approaches through both an automatic and human evaluation and find that modular difficulty control is an effective method for constraining LLMs.
Location
This talk was recorded on February 26th, 2025 as part of the EACL 2026 conference in Rabat, Morocco.