Exploring interdisciplinary boundaries: A BERTopic modeling approach to analyze four interdisciplinary fields at the abstract and sentence levels

Interdisciplinary research plays a central role in addressing complex global challenges; however, its linguistic realization remains underexplored. This comparative study investigated the thematic structures of interdisciplinary academic writing using BERTopic, a context-sensitive topic modeling method, which leverages sentence embeddings. Drawing on a corpus of 4,000 research abstracts from four mature interdisciplinary fields – biomedical engineering, environmental science, cognitive science, and digital humanities – this analysis compared topic structures at two granularity levels: abstracts and individual sentences. At the abstract level, six dominant topics exhibited a strong disciplinary alignment, indicating that even in explicitly interdisciplinary domains, boundaries remained prominent. However, sentence-level modeling revealed a more complex landscape with 80 fine-grained topics, which included both discipline-specific and multi-disciplinary categories. The latter were classified into shared research focus, methodological convergence, and research practice. These findings suggested that, while interdisciplinary writing retained core disciplinary identities, it exhibited selective semantic integration across fields. This study further argued that mature interdisciplinary fields served as coherent units for comparative discourse analysis. It contributed to the interdisciplinary discourse and offered implications for ESP research.