Causality extraction system detects false cause claims in news accurately
Shrome at Touché: Soft-Vote Ensembling and Counter-Causal Augmentation for Causality Extraction
Computation and Language
Summary
News stories sometimes say something caused something else, but sometimes these statements deny the cause. The authors developed a system that can tell if a sentence really claims causation, find the cause and effect parts, and decide if the claim supports or rejects the cause idea. They used multiple models working together and added examples created by a large language AI to train the system better for tricky cases. Their system scored well in tests, especially in accurately finding cause and effect in sentences.
What this means in practice
- •For news monitoring teams: Detect and analyze sentences in news articles that falsely suggest causes to improve misinformation tracking.
- •For fact-checking services: Improve automated tools to flag statements that deny causal relations despite causal language, enhancing fact-check accuracy.
Authors
Roham Zendehdel Nobari, Shayan Sooratgar
Abstract
Touché 2026 extends causality extraction to counter-causal claims: news sentences whose surface form appears causal but whose meaning denies the causation, as in "It is falsely believed that X caused Y." A system that relies on surface cues such as "caused" or "led to" will accept such a sentence as causal and give it the wrong polarity. On the Countercausal News Corpus (CCNC), the task has three subtasks: deciding whether a sentence is causal (detection), locating its cause and effect spans (extraction), and labeling its polarity as procausal, counter-causal, or uncausal. We build one model per subtask. Detection is a fine-tuned classifier with a single cross-task rule that uses the extracted spans to remove false positives. For extraction, we ensemble three RoBERTa-large BILOU+CRF taggers by averaging their token-level scores before decoding, rather than voting on the spans each tagger produces. For polarity, where labeled counter-causal examples are scarcest, we add training sentences generated by a large language model prompted with nine patterns of counter-causal expression adapted from Hagen et al., keeping only those that pass automatic structural checks. On the held-out CCNC test set, the system reaches F1 0.869 on detection and macro-F1 0.817 on polarity, and in the organizers' final causal-only evaluation of extraction it scores granularity-adjusted F1 0.728, the highest extraction score among all submissions including the organizers' baseline. The development split is used only for component selection and the ablations reported in the paper.