AI System Learns to Write Dictionary Definitions Optimized for Language Learners

Writing dictionary definitions—especially ones suitable for language learners—demands significant time and expertise. A new research effort explores whether machine learning models can automate this lexicographic task while maintaining quality.

The Challenge of Automated Definition Writing#

Creating definitions for learner's dictionaries presents a particular challenge: words must be explained using accessible vocabulary that won't confuse students. Researchers framed this as learner's dictionary definition generation (LDDG), where an AI system must produce clear, non-technical explanations for headwords.

Building an Evaluation Framework#

Before tackling the generation problem itself, the team first needed a reliable way to judge whether definitions are good. Working with a professional lexicographer, they developed:

• New evaluation criteria specifically tailored to dictionary definitions
• An LLM-as-a-judge approach, leveraging language models to assess quality
• A reference dataset of Japanese dictionary definitions

Testing revealed their evaluation method achieved reliability levels matching human agreement, suggesting it can serve as a trustworthy benchmark.

The Simplification Strategy#

The researchers proposed an LLM-based system using iterative simplification—a technique where the model repeatedly refines definitions to use progressively simpler vocabulary. Results indicate this method produces definitions scoring well on their criteria while maintaining strong lexical simplicity, a key requirement for learner resources.

Why It Matters#

Automating definition creation could dramatically reduce the cost of producing educational dictionaries, making quality language learning resources more accessible across languages and skill levels.