arXiv cs.CL· Junwei Ma, Arun Kumar Chithanar, Palash Goyal, Chul Lee·· 11 小时前AI 评分31
表格序列化方式决定语言模型保留哪些关系
Table Serialization Determines Which Relations Language Models Preserve
AI 导读
研究显示，指令微调语言模型在表格查找中的表现取决于文本序列化方式而非表格形状本身：在 256 单元格合成表格上，行优先格式在 64 x 4 高表中领先 42.3-54.0 个百分点，列优先格式在 4 x 64 宽表中领先 35.0-59.0 个百分点。错误模式呈互补性：行优先易保留正确行但丢失目标列，列优先易保留正确列但丢失行，作者称之为“关系局部性”。
正文 · 原文
Abstract:We study how text serialization affects table lookup in instruction-tuned language models. Our diagnostic asks the model to find the row containing a specified value under one named column, then return the value under another named column in that same row. We render the same synthetic 256-cell tables in row-major form and in column-major form with each header placed beside its values, while varying the table aspect ratio across five shapes. Across three models, the better format changes with table shape: row-major performs better on tall tables, whereas column-major performs better on wide tables. At 64 x 4, row-major leads by 42.3-54.0 percentage points; at 4 x 64, column-major leads by 35.0-59.0 points. Mapping wrong answers back to their source cells reveals a complementary error pattern: row-major tends to preserve the correct row but lose the target column, while column-major tends to preserve the correct column but lose the row. We use the term relational locality to describe this pattern: serialization keeps one relation locally available while making the other depend on positional alignment. A separate non-tabular ordering experiment supports this interpretation. Thus, apparent width-related failures in text-serialized tables depend on linearization, not table shape alone.
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.24798 [cs.CL] |
| (or arXiv:2610.24798v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.24798 arXiv-issued DOI via DataCite |
Submission history
From: Junwei Ma [view email]
[v1]
Sat, 17 Oct 2026 18:07:24 UTC (83 KB)
[v2]
Thu, 29 Oct 2026 22:30:05 UTC (83 KB)
来源：arXiv cs.CL · arxiv.org