Disciplinary Gaps in Subject Indexing: A Structural Analysis of Controlled Vocabularies’ Breadth and Depth

This chapter delves into the structural analysis of controlled vocabularies, focusing on their breadth and depth, and their impact on subject indexing performance using large language models (LLMs). The study examines how these dimensions vary across ten disciplines, revealing significant differences in indexing accuracy. Key findings include the positive correlation between breadth (coverage) and indexing performance, and the negative correlation between depth (structural complexity) and performance. The chapter introduces a framework to quantify these dimensions, providing practical insights for designing more effective and equitable indexing systems. Additionally, the study highlights the importance of balancing terminological coverage with semantic granularity to enhance indexing accuracy. The results offer valuable guidance for vocabulary curation and system design, emphasizing the need for discipline-sensitive approaches in the age of large-scale language models.