Getting Sick After Seeing a Doctor? Diagnosing and Mitigating Knowledge Conflicts in Event Temporal Reasoning
This addresses a specific issue in natural language processing for event understanding, with incremental improvements in mitigating model biases.
The paper tackles the problem of knowledge conflicts in event temporal reasoning, where models' prior biases conflict with actual event relations in narratives, and proposes a Counterfactual Data Augmentation method that reduces hallucination and improves performance for both Pre-trained and Large Language Models.
Event temporal reasoning aims at identifying the temporal relations between two or more events from narratives. However, knowledge conflicts arise when there is a mismatch between the actual temporal relations of events in the context and the prior knowledge or biases learned by the model. In this paper, we propose to detect knowledge-conflict examples in event temporal reasoning using bias indicators, which include event relation prior bias, tense bias, narrative bias, and dependency bias. We define conflict examples as those where event relations are opposite to biased or prior relations. To mitigate event-related knowledge conflicts, we introduce a Counterfactual Data Augmentation (CDA) based method that can be applied to both Pre-trained Language Models (PLMs) and Large Language Models (LLMs) either as additional training data or demonstrations for In-Context Learning. Experiments suggest both PLMs and LLMs suffer from knowledge conflicts in event temporal reasoning, and CDA has the potential for reducing hallucination and improving model performance.