用反事实提问提升大模型的时间推理一致性
Counterfactual-Consistency Prompting for Relative Temporal Understanding in Large Language Models
- 设计反事实提示，通过生成反向问题增强时间逻辑约束
- 在多个数据集上显著改善显式与隐式事件排序准确率
- 适合需要严谨时间推理的问答与内容生成任务
尽管大语言模型具备强大能力，其时间推理性能仍不理想。现有研究指出，模型在理解事件时间关系时易出现矛盾，例如混淆‘先于’与‘后于’等互斥关系并做出不一致判断。本文提出一种新型反事实提示方法，通过生成反事实问题并施加联合约束，强化模型的时间一致性。在多个数据集上的评估表明，该方法能有效解决时间不一致问题，在显式与隐式事件排序及时间常识理解任务中均取得显著提升。
原文摘要 · Abstract (English)
Despite the advanced capabilities of large language models (LLMs), their temporal reasoning ability remains underdeveloped. Prior works have highlighted this limitation, particularly in maintaining temporal consistency when understanding events. For example, models often confuse mutually exclusive temporal relations like ``before'' and ``after'' between events and make inconsistent predictions. In this work, we tackle the issue of temporal inconsistency in LLMs by proposing a novel counterfactual prompting approach. Our method generates counterfactual questions and enforces collective constraints, enhancing the model's consistency. We evaluate our method on multiple datasets, demonstrating significant improvements in event ordering for explicit and implicit events and temporal commonsense understanding by effectively addressing temporal inconsistencies.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。