Zero-shot event causality identification using multi-reasoning task question answering with large language models

Event Causality Identification (ECI) is a crucial task in Natural Language Processing (NLP) that focuses on identifying causal relationships between events in textual data. Despite recent advancements in few-shot learning, Large Language Model (LLM)-based ECI models often encounter causal hallucinations, leading to high recall but low precision. To address this issue, we propose MuTQA, a zero-shot ECI model that utilizes multi-reasoning task question answering. MuTQA incorporates four causal reasoning tasks—forward causal reasoning, masked causal reasoning, counterfactual reasoning, and contrapositive reasoning—using a fusion strategy based on consistency and confidence. Our experiments show that MuTQA outperforms traditional zero-shot models and LLM-based models, particularly in improving precision and minimizing errors caused by causal hallucinations.