Search papers, labs, and topics across Lattice.
The paper introduces ClimateCause, a new expert-annotated dataset of complex causal relationships extracted from climate science reports, focusing on implicit and nested causal structures. The dataset normalizes cause-effect expressions into individual causal relations, annotated with correlation, relation type, and spatiotemporal context. Benchmarking LLMs on the dataset reveals that causal chain reasoning remains a significant challenge, despite progress on correlation inference.
LLMs still struggle to reason through causal chains in climate science, even when they can identify individual correlations.
Understanding climate change requires reasoning over complex causal networks. Yet, existing causal discovery datasets predominantly capture explicit, direct causal relations. We introduce ClimateCause, a manually expert-annotated dataset of higher-order causal structures from science-for-policy climate reports, including implicit and nested causality. Cause-effect expressions are normalized and disentangled into individual causal relations to facilitate graph construction, with unique annotations for cause-effect correlation, relation type, and spatiotemporal context. We further demonstrate ClimateCause's value for quantifying readability based on the semantic complexity of causal graphs underlying a statement. Finally, large language model benchmarking on correlation inference and causal chain reasoning highlights the latter as a key challenge.
