当LLMs粗心读取表格时：数据引用错误的测量与减少

摘要

尽管大型语言模型（LLMs）在表格任务中表现良好，但它们仍会出现数据引用错误（DREs），即在理解表格结构的前提下，错误引用或遗漏表格中的数值。除了影响最终答案的准确性外，DREs还会直接损害中间推理步骤的正确性和可靠性。然而，以往的研究仅提供了有限的小规模分析。在本工作中，我们首次对不同模型和任务中的表格数据引用错误进行了系统评估。结果表明，所有被测模型（参数量从1.7B到20B）均存在DREs。此外，我们证明，通过基于评判模型的过滤和拒绝采样，将数据引用作为评判标准能显著提升答案准确率，最高提升12.0%。最后，我们训练了一个轻量级4B参数评判模型，在检测分布内和分布外DREs时，平均F1分数达到78.2%，并能有效辅助更大模型的推理过程。

English

While large language models (LLMs) perform well on table tasks, they still make data referencing errors (DREs), i.e., incorrectly citing or omitting table values, despite understanding the table structure. Beyond final-answer accuracy, DREs directly compromise the correctness and reliability of intermediate reasoning steps. Yet prior studies have only offered limited, small-scale analyses. In this work, we present the first systematic evaluation of tabular data referencing errors across different models and tasks. Our results show that DREs occur across all tested models (1.7B to 20B parameters). Furthermore, we demonstrate that incorporating data referencing as a critic significantly improves answer accuracy up to 12.0%, through critic-based filtering and rejection sampling. Finally, we trained a lightweight 4B-parameter critic model that achieves an average F1 score of 78.2% in detecting both in-distribution and out-of-distribution DREs, and effectively assists inference for larger models.