---
id: 20260505-T0-10
title: "RSAT：让小模型学会“引用”单元格，表格推理查得清来源"
title_en: "RSAT Trains Small Language Models to Cite Sources in Table Reasoning"
url: https://ai.daily.yangsir.net/daily/20260505-T0-10
issue_date: 2026-05-05
publish_date: 2026-05-04T04:00:00.000Z
category: research
source_name: "arXiv cs.CL (NLP)"
source_url: https://arxiv.org/abs/2605.00199
---

# RSAT：让小模型学会“引用”单元格，表格推理查得清来源

语言模型在处理表格问答时，往往难以验证其回答究竟参考了哪些数据单元格。为了解决信任和溯源问题，研究人员提出了RSAT方法。该方法专门训练参数量在1B到8B之间的小型语言模型（SLM），使其在回答问题时能够生成带有单元格级别引用的逐步推理过程。这种结构化归因机制使得用户可以精准核查每一步推理的数据来源。开发者可以用它来构建低成本且透明度更高的数据分析应用。

## English Version

**RSAT Trains Small Language Models to Cite Sources in Table Reasoning**

Language models often struggle to verify which table cells inform their answers. Researchers introduced RSAT, a method training small language models (1B-8B) to generate step-by-step reasoning with cell-level citations. This structured attribution allows users to trace answers directly to source data, enhancing transparency.

---

**来源**：[arXiv cs.CL (NLP)](https://arxiv.org/abs/2605.00199)

**详情页**：https://ai.daily.yangsir.net/daily/20260505-T0-10

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*