Evaluating Large Language Models for Standard Names Generation in CF Conventions

This presentation was given at the CF Workshop 2025 (22–25 September 2025). It explores the use of Large Language Models (LLMs) to support the CF community in generating and managing standard names. The work evaluates two prototype tools: a generator that creates candidate CF standard names from textual descriptions, and a semantic search tool trained on CF vocabularies to suggest existing names matching a user’s description. The talk discusses potential benefits for improving consistency, scalability, and accessibility of the CF conventions, while also highlighting limitations and shortcomings found in the current standard name table that affect model performance. Future directions include improving training data quality, refining model precision, and transforming these prototypes into practical tools for the CF community.