Learning Implicit Editorial Rules for Medicinal Terminology Naming Using Rule-Compositional Cascading Prompts

Area of Science:

• Medical Informatics
• Natural Language Processing
• Ontology Engineering

Background:

• Medicinal terminologies rely on expert-defined names (FSNs, PTs) with implicit editorial rules.
• These rules are often not formally encoded in existing ontologies.
• Automating the generation of accurate and consistent medicinal terms is a significant challenge.

Purpose of the Study:

• To explore the use of large language models (LLMs) for automated medicinal terminology naming.
• To learn and apply implicit editorial rules governing FSN and PT generation.
• To evaluate different LLM approaches for improving terminology consistency and accuracy.

Main Methods:

• Utilized existing OWL-defined concepts from the HSE Ireland medicinal terminology.
• Evaluated a four-level approach: direct fine-tuning and rule-compositional cascading prompts.
• Applied methods across six distinct medicinal concept types.

Main Results:

• The cascading prompt framework demonstrated improved accuracy and numerical robustness.
• This approach enhanced interpretability by explicitly modeling editorial decision steps.
• Scalable and consistent generation of FSNs and PTs was achieved.

Conclusions:

• LLMs offer a powerful solution for automating medicinal terminology naming.
• Cascading prompt frameworks significantly enhance the reliability and consistency of automated naming.
• This work enables more efficient and standardized medicinal terminology management.