A Comparative Linguistic Analysis of the 2014 and 2023 IPCC Synthesis Reports using Topic Modeling and Scattertext

The study examines the thematic and lexical variation between two consecutive assessment reports (2014 and 2023) from the International Panel on Climate Change (IPCC). The synthesis sections of both reports were analyzed via three of the most widely recognized topic modeling techniques: BERTopic, NMF, and LDA. To enhance and validate the topic modeling results, the Scattertext tool was also employed. The most coherent and interpretable output was produced by NMF, which yielded valuable terminological insights. The 2014 report primarily focuses on topics such as the anthropogenic causes of climate change, radiative forcing, modeling of future rise in greenhouse gas concentration levels, and sea level rise. In contrast, the 2023 report emphasizes climate change finance and governance, integrating climate mitigation into the broader context of sustainable and resilient development. It also introduces new concepts such as socioeconomic pathways, net zero, carbon budgets, climate resilience, and adaptation gaps. While topic modeling is typically used for single-text analysis, we show that is can be successfully applied for comparative and, more specifically, diachronic research in linguistics. Additionally, we demonstrate how scholars in the fields of climate discourse and terminology can benefit from the integration of NLP tools. The article discusses the challenges of applying topic modeling to jargon-heavy, homogeneous texts, as well as the preprocessing and fine-tuning requirements of BERTopic, NMF, and LDA, each based on a different algorithm.