| Sumario: | As scientific consensus on anthropogenic climate change has solidified, outright denial in political debate has largely given way to more subtle forms of climate change rejection. These so-called climate delay discourses acknowledge the problem yet justify inaction, posing a growing challenge to effective climate communication and policymaking. This study operationalises Lamb et al.'s (2020) typology of climate delay discourses to examine 25 years of Danish parliamentary speeches (1998-2022). Using a keyword-based retrieval method, we identified approximately 34,000 climate change-related speech segments and applied large language models (LLMs) to classify them according to delay discourse categories. We compared zero-shot and few-shot prompting strategies, including variations with chain-of-thought reasoning, to evaluate LLM performance on complex rhetorical classification tasks. Few-shot prompting delivered promising results with respect to both recall and accuracy, while chain-of-thought reasoning provided limited benefits and, in some cases, harmed performance. Temporal and partisan analyses reveal that delay discourses have been consistently present in Danish political debate, with a marked increase in recent years. The most prevalent discourse, all talk, little action, reflects the gap between ambitious climate targets and policy implementation, particularly among governing parties. While right-leaning parties often shift responsibility away from Denmark, left-leaning and green parties more frequently invoke appeals to social justice. Our findings demonstrate both the promise and the limitations of LLMs for large-scale political discourse analysis and provide evidence that climate delay discourses are a routine part of Danish parliamentary debates.
|
|---|